Drop inconsistent odometry

[truher]

when one wheel is slipping a lot relative to the others, don't include it in the average.

note since the odometry uses differentials, the "bad" wheel needs to have a "fake" odometry created for it.

[truher]

note that wheel slip is less of a problem with velocity control, assuming we're operating inside the feasible envelope. so for now this is less important. if we switch to torque control, this will be a much bigger deal.

[truher]

to solve it:

• solve chassis speeds or deltas for four wheels
• project this solution back into the expected module states
• compare the actual module states to the expected module states
• wheels that are going "too fast" should be ignored
• to ignore them, solve chassis speeds again using a different kinematics matrix that drops one or two wheels.