Medical grade power supplies have supposedly been tested to specific standards. In particular, the maximum leakage current between the hot and isolated sides will be very low. There are also other internal requirements, like a large creapage distance accross the isolation barrier.
The medical-grade requirements are to make the supply more safe, even if the isolated side is connected to a human more closely than thru ordinary skin. Some of the requirements are to reduce the chance of failures, and to decrease the chance that any one failure could electrocute someone.
All this costs money, so medical grade power supplies are significantly more expensive than ordinary ones. The volumes are less, which again drives the price up.
These extra specs have nothing to do with voltage stability, ripple, current rating, and the like.
Ah yes, the LT1054. A part apparently designed to separate engineers who read the datasheet from engineers who don't. Congrats on being on the winning team!
So, the problem is not as bad as you think. The VOUT pin only needs to not be pulled positive during startup. Once the LT1054 has started, loads may be powered from it without worrying if the current returns to GND or a positive rail.
The transistor circuit they show in the datasheet simply acts as a switch that can only be turned on by the emitter being pulled to a voltage lower than the base. Since the base is grounded, the VOUT pin is effectively disconnected from the load until it has started and is pumping charge and pulling VOUT below ground. So the collector of the transistor is effectively your new output pin, but you may tie all loads you will be powering to the collector, it doesn't need to be done per load, but once per LT1054.
There will be no problems regardless of the number or nature of the loads powered from the LT1054's negative rail (the collector of the BJT), as the BJT is a complete cure.
As for the AD8275 and AD7176-2, they will work fine. The AD8275 is a good choice and does exactly what you need - it will take a bipolar large swing signal (±10V?! What an inconvenient sensor output level) and turn it into a unipolar downscaled signal without even needing a negative rail. They're designed to work with 5V and GND, so as long as their GND is the same as the LT1054's and everything else's GND, then it's irrelevant as to what their power source is. Whether its 5V derived from your 12V rail with a simple LM7805, or some crazy isolated DC/DC module, or two 3V coin cells with an LDO dropping it to 5V, all will work just as well. As long as the 5V power supply's ground is the same ground as everything else. Ground is ground. It is what the voltages are in reference to, so 5V is only 5V in reference to something else. For a battery, it's the negative terminal. If you connect the positive terminal of a battery to a circuit but not the negative terminal, then there is not 5V of potential anywhere on the circuit, because the battery's 5V of potential is only with respect to the battery's negative terminal, and simply 0V when referenced to whatever part of the circuit it is connected to. However, connect the negative terminal to the right spot, and suddenly there is a voltage potential across the circuit. Connect another battery at a different voltage, and whatever you connect it to will be at that potential. If you connect it's negative terminal to the ground that is shared with the circuit and other battery, then it's negative terminal is at the same potential as GND, and its positive terminal is however many volts above that as the battery produces. OR, if you wish, connect the POSITIVE terminal of the battery to ground, and now the positive terminal is at that potential. That means the negative terminal is however many volts BELOW the ground potential. A trivial way to get, say, ±9V for prototyping is simply two 9V batteries in series. It might look like 18V, but its also ±9V if you chose GND to be the terminal where the + of one connects to the - of the other.
The important thing to understand here is that voltage potentials are only in reference to something else, and relative. "Ground" is completely arbitrary. It's a label, nothing more. Ground is whatever you decide it is. You can just as easily take your circuit and say -12V is ground, and that the sensor is running from ground and 24V. The AD8275 is powered by 12V at it's VSS- and 17V at VSS+. This is just as valid and correct, and altering these labels does not alter the circuit or make it cease to function. It just makes it a lot harder to think about, so a potential 12V above the lowest potential is chosen as ground. Again, it's all just arbitrary labels. You can even chose an imaginary -100V and reference everything to it. GND in your circuit becomes.... -88V, or 12V above -100V. This would be silly and pointless of course, but I think it makes my point.
The sooner you start thinking in these terms, the easier all this power rail nonsense becomes, and you won't worry about having multiple voltages anymore. Just pick what you want to be ground, and reference everything to that, then you won't have any problems.
This was way more information than you asked for, but I got the feeling this was all stemming from not quite getting the actual nature of voltage potentials (which takes many engineers years if they do at all, it's subtle, so don't feel bad). Without that grasp, all these different voltages and dual polarities and parts with weird little gotchyas about pulling the pin above GND etc. can all get very confusing very quickly, but there is a simple, elegant way of thinking that makes it all melt away.
If you're more of a visual learner, the metaphor that fits perfectly is height and potential energy. Voltage is simply how high something is. Ground is...well, the ground. Drop rocks from different heights, they all fall to the same ground on the same planet from the same gravity. Dig a ditch, and its below ground, a negative potential. It takes energy to move something out of the ditch up to ground, where as above ground, you simply drop it. Conversely, rocks already on the ground can fall further into the ditch. The LT1054 is a guy constantly digging a ditch but he only stops digging once it's a certain depth, then only shovels out as much dirt or rock as gets thrown into his ditch. However, if rocks start being dropped from higher up into his ditch area before he's even gotten a chance to dig the hole (like if the VOUT pin is pulled positive), this uh pisses him off so he refuses to start working. The metaphor doesn't extend to latch up, sorry.
Hopefully this long ramble helped, or at least didn't make you even more confused.
Assuming you're doing the sort of thing that can live without galvanic isolation, there are plenty of easy solutions. If you don't mind a transformer, here is one of the lowest parts-count (and probably lowest cost) solutions, using the ubiquitous LM2577:
The simpler solutions (as with the cheaper DC-DC converters) tend to have so-so regulation, require a minimum load etc. Be mindful of this if you're up near the absolute maximum voltage input of the load semiconductors- the output voltage could be volts above nominal with a very light load.