GRP Shadow Set Issues with the Attitude Estimator

[kylekrol]

Sometimes the GRP attitude representation error just explodes with the attitude estimator. @stewartaslan mentioned that it may be because of a "shadow set" of GRP values. Not really sure what this means at the moment but negating the initial equation of the dynamics simulation solves the issue. To be clear, the filter actually doesn't perform well but it's GRP representation doesn't agree with the truth model - as in a negative sign is flipped but the estimated attitude is still correct.

The results can be currently replicated with the following command:

python -m psim -s 1000 -p fc/attitude -ps 1 -c sensors/base,truth/base,truth/detumble -v AttitudeEstimatorTestGnc

on master with this commit 3d6c340.

Some plots are given below. As you can see, the GRP error is horrid but actual degrees error and angular rate estimation error are still performing very well. The gyroscope bias estimate plots are not shown here but are behaving nominally.

[kylekrol]

Marking this an noncritical given the results of #319.