Fixes #2806 (I Hope). I am merging this in so RO people may be able to test it. I have tested it about as well as I can without RO installed.

The change is mainly that SteeringManager now has an epsilon error amount where it "doesn't care" about the rotational rate being a bit off until later on when it gets closer to the target orientation. If you're trying to rotate at 1 degree per second toward the desired orientation and instead you're rotating at 1.05 degrees per second toward the desired orientation, you shouldn't really care yet about that small difference (and not bother correcting it with precious RCS fuel) until you're very close to the destination orientation.

This is accomplished by adding an epsilon to the PIDLoop structure, then using that epsilon in the PIDLoops the SteeringManager uses, and having the size of epsilon "Lerp" between a max and min value (max when pointed wrong where precision doesn't matter yet, and min when pointed right.)

The default values for these min and max epsilons are chosen arbitrarily and could be adjusted in future PRs if people find them problematic in testing.

While I was in there I also noticed that the existing EPSILON used in the torque pids was so small as to be utterly meaningless, and I increased it a bit, (While also splitting it into two constants because there were two different unrelated things the same EPSILON constant had been used for before, one of which should remain small and the other should not.)

There is also a lot of documentation to go with the changes. If you are trying to test this out, read through the documentation diffs on this PR to see what there is to play with.