I wouldn't worry about losses from the microcontroller's voltage regulator. Yes, most of the energy goes into making heat. But, the current needed by the microcontroller will be so small compared to the motors that it's probably not worth concern. For more detail on how to calculate just how much heat there will be, see My linear voltage regulator is overheating very fast.
I would suggest dropping the voltage regulator for the motors. A PWM-driven motor is already a buck converter, so as long as you aren't exceeding the maximum voltage of the driver, you are probably safe. The voltage rating for motors is usually the maximum continuous voltage they can take, but peak voltage is much higher, limited basically only by the insulation in the windings. As long as your motor controller is doing its job of limiting the current, and thus the power and heat, to the motor, a higher voltage motor drive is fine. Adding a regulator to the system just makes it more inefficient.
The motor will generate noise. Reduce it by including plenty of bypass capacitors, from small ones around each IC, to big ones to supply the motor. Also, arrange your PCB and cables to minimize the area of the loop through which the motor current travels. This will minimize the inductance of that loop, and thus reduce its inductive coupling to everything else in your circuit. Remember that the motor current flows not only through the positive supply but also ground. Keep these currents away from the microcontroller's ground. Your motor is small enough that you should not require any extraordinary measures beyond good layout and standard practice to keep noise at reasonable levels. I have a previous answer on noise with some more detail. Also, a linear regulator usually has better power supply noise rejection than a buck converter: another reason to retain the 5V regulator.
I'm not enough of an expert on batteries to address your concerns beyond basic advice, like include a fuse. This seems like something you could easily break out into a separate question and get some good advice, if after doing some basic research you require more clarification.
There is, in a sense, no qualitative difference. The difference is one of scale, both of current and of time.
A bulk capacitor is used to prevent the output of a supply from dropping too far during the periods when current is not available. For line-powered linear supplies, this would occur during the periods (say, 10s of msec) that the line voltage is near zero. It also applies to the circuit as a whole. That is, an electronics assembly containing multiple circuit cards might have a single set of bulk capacitors in the power supply.
Decoupling capacitors, on the other hand, are used locally (such as 1 per logic chip in some systems) and are intended to supply current for much briefer periods (typically 10s of nsec for TTL systems) and much smaller currents. As a result, decoupling caps are typically much smaller than bulk caps.
This is not entirely a hard and fast rule - for some high-speed analog parts a mix of different decoupling values is recommended, with the smallest values providing the shortest compensation times, and larger caps being used as well. High-speed A/D converters often used to recommend a 0.1uF / 10 uF combination. Many logic boards have a mix of values scattered around. CPUs, in particular, are often surrounded by largish (10 - 100 uF) electrolytics, with a whole bunch of small SMD ceramic caps right under the chip.
As for demonstration circuits, only bulk caps make easy demo's. Take a transformer output of, let's say, 6 VAC, and run it through a bridge rectifier. Load the output of the bridge with a power resistor (like, 10 ohms) and look at the voltage across the resistor - it will drop to zero 120 times per second (100 if your line frequency is 50 Hz). Now place a bulk cap of 10,000 uF on the bridge output, and the output will be much smoother, with 120 Hz dips - it will look sort of like a sawtooth - but in general the voltage will be much smoother.
Decoupling is harder. Try setting up an op-amp amplifier on a solderless breadboard using a high-speed op amp and long wires running from the breadboard to the power supply. There's a good chance the output will oscillate with no input. If you put 0.1 uF ceramic caps from the supplies to ground, and do it right at the op amp supply pins, this will often clear up the problem. Or not - solderless breadboards aren't good for high-speed work even if you're careful, and some op amps are very stable, but it's the best suggestion I can come up with.
Best Answer
You need at least to know how much supply ripple your different components will tolerate. This is your starting point.
Then I would draw an equivalent circuit of your power distribution network and the main load (the motor as an inductor in series with a resistor and a switch)
simulate this circuit – Schematic created using CircuitLab
You will need to approximate/measure the L and R values for the cables going to the H-Bridge and other parts. Do a transient simulation where you open the switch after the motor has reached maximum current. You should see a large oscillation at the voltage input to the R-Pi. Adjust C2 and C3 to reduce it to acceptable levels.
If you cannot measure the inductances, try approximating them using an online calculator.