This is a tall order. Let me recap what I understand from the original question as well as comments made:
You have about 500 devices, each costing about US$200 each. You want to protect them from being stolen from an office park, and ideally there would be a tracking system so they could be recovered. The solution shouldn't increase the per-unit cost much, either in initial cost or recurring (monthly) costs.
Let me start my answer by telling you what won't work...
GPS: You didn't say what these devices are, but I am going to assume that they will be indoors almost all of the time. GPS reliability indoors is almost zero. If you are careful you could get one to work, but metal and concrete effectively block GPS signals. Wood does block the signal, but not as effectively as metal and concrete. The point is, GPS is not going to be the ultimate solution for your location requirements.
ZigBee, Bluetooth, Wireless: None of these are reliable enough to use for your communications. They would be OK within your office park, but not outside of it where you are essentially hoping that there is an unsecured network that you can use.
Pager/Cell-phone: From a technological point of view these would probably be your best solution, but the monthly costs will be prohibitive. As @kenny said, US$15/month would be the lowest price.
Battery Life: With any of these solutions you will be very limited by battery life. You might get a couple of days, max, and that assumes that you have enough space to put a reasonable size battery in it and the stuff required for 500 people to charge it.
So, here are some "out of the box" ideas that might work for you. If it doesn't work then maybe it will spark an idea or two.
Use WiFi, but not for tracking. When the unit powers up it tries to establish a Wifi connection with the office park network. If it can't then the unit simply shuts down. In this way, if the unit is taken out of the office park it just doesn't work. While this doesn't do any tracking/recovery it does reduce the motivation for theft. After boot, the WiFi module can be powered down thus saving battery life.
Use medium or long range RFID to detect the unit leaving the park. For example, place some RFID readers at the doors to buildings and log when a unit goes through. Combine this with security camera video and you can identify who took the unit. The huge advantage with this is that it requires no batteries and the cost per unit is very low. The initial investment might be high, but after that it doesn't really matter if your tracking 500 units or 5,000 units. And since it's RFID, you can protect just about anything without a huge engineering effort.
Hope that helps!
EDIT:
New info: Units will be mostly outside. Location tracking is most important within the office park/campus. Theft prevention isn't top priority.
In that case, I'd just use WiFi for communications with or without a GPS for location. Maybe include a GPS but use WiFi triangulation for when GPS reception isn't quite working (since you already have a WiFi module). If you find out that WiFi triangulation is good enough then you can skip the GPS on later units.
The main issue I see with this is the expense of the WiFi & GPS stuff. I'm estimating that it would cost about US$75-100 for such a circuit, which is not an insignificant fraction of the unit cost. I don't really see an alternative that would be as practical.
I still don't think ZigBee or Bluetooth is reasonable. It would work, but would require extra infrastructure. With WiFi, you can at least share the costs with the I.T. department.
Generally if you can't find the part, you'd have to make it yourself. There are many tutorials explaining how to make a part in Eagle.
You are correct in discerning between the three categories for an Eagle part.
Schematic symbol:
This is the representation of your part as it will appear in the schematic. It doesn't have to have all the pins represented exclusively, like say if there are multiple GND pins.
Footprint/package:
This is what the devices physically looks like and how it will appear on your circuit board. Each solderable pin must be represented here. You need this so you can actually place the part on your PCB.
Device:
This is the middle ground between the two, while you never place a device, it is how the schematic symbol will connect to the foot print in the software. You need this because a connection made in a schematic represents a physical connection on the circuit board. If you connect a signal to a pin of a symbol in the schematic editor, the software needs to know to make the same connection on the circuit board.
You'll notice, after you've made a schematic that has connections, that when you start to design your circuit board, you'll see many yellow lines, these are called "air wires" and are un-routed representations of the signals created in the schematic. In the PCB editor, you are just making the connections you made in the schematic editor a physical track that will be on your circuit board when you have it fabricated.
For this part specifically, I know this has been made and released since I use it all the time and I didn't make my own. Sparkfun releases many of the parts used in their products in the form of the footprint and schematic symbol. It seems like they have a DIL28 connector in their connections library. You can download them from the linked github page.
As for more Eagle library resources:
Best Answer
To directly answer your question, "Is anyone producing discrete memristors?" the answer is no. But as you discovered in the linked article, there are ways to make your own memristors from objects that are almost trash, so let's discuss how one observes what they do, and what it is that defines a memristor.
I think the most telling image from the linked article is this one, which I've annotated with numbers for discussion:
The article says:
The curve tracer is simply a device that applies a varying voltage to some device under test and measures the current, or applies a varying current and measures the voltage, and then graphs current on one axis and voltage on the other. One can be made with an X-Y scope and a function generator:
simulate this circuit – Schematic created using CircuitLab
V1 is the output of the function generator. Channel A on the scope is mapped to the horizontal axis, and measures voltage. Channel B is mapped to the vertical axis, and measures current indirectly as the drop over R1, which is selected to be small enough to have negligible effect on the device under test (DUT).
Here's what's happening. Let's start at the origin (1), at \$0V\$ and \$0A\$. Initially, the memristor is in a high resistance state, so as the curve tracer applies a higher voltage, not much current develops (2).
At some point, enough voltage has been applied long enough. The time integral of voltage is flux, even though there may not be any magnetic flux building in the component, as there would be with an inductor. I'm an engineer, not a physicist, but I'm guessing that the applied voltage reduces the corrosion on the brass, leading to a better, and less resistive contact with the aluminium. Now that the resistance is less, some current can develop (3).
Now the curve tracer is reducing the voltage, and we get a corresponding decrease in current, until we get back to \$0V\$ and \$0A\$ (1)
The curve tracer continues to reduce the voltage to negative. We see a negative current develop, along with some noise (4). The decreased slope of the line from (1) to (4) suggests the resistance was higher here than it was from (3) to (1). I suspect this is because the negative voltage is oxidising the corrosion between the brass and the aluminium, increasing its resistance. Or maybe there's some rectifying property in play here, similar to a point contact diode.
At some point, we've applied enough negative voltage long enough that the time integral of voltage (flux) has gone negative enough to get the memristor back to the high-resistance state, and the current decreases to almost nothing (5).
The curve tracer increases the applied voltage from the maximum negative value towards \$0V\$, until we get back to (1), and the cycle repeats.
Viewed this way, one could say the important properties of a memristor are:
The example curve above shows a lot of noise and nonlinear behavior, probably owing to the less than carefully engineered nature of the setup. Here's a more idealized version of the same thing:
(from Electrical Charging of Memristors)
This shape is pretty typical, but according to most definitions of memristor, this particular shape isn't required, as long as there is a pinched hysteresis loop that passes through \$0V, 0A\$. Consequently, there isn't just a scalar number that describes an "ideal" memristor as there is with resistors, capacitors and inductors. Rather, the slope of this curve is specified as a function, in units of volts per amperes, or ohms.