Fix Pruning

[varunagrawal]

Update the return value of pruned factors/conditionals to return ∞ as the error in negative log space.

This way, when we normalize (by subtracting the minimum value) and take the exponential negative exp(-e), we get p=0 for pruned assignments, valid probabilities everywhere else.

With this fix, we can now run hybrid legged state estimation for 16 timesteps in 3.8 seconds, down from 13 minutes.

[dellaert]

Very cool. Might want to update the PR comment with what you told me in email. Also, a test that fails before but works after would be great, so future PR’s don’t mess this up.

[varunagrawal]

Added a test which originally fails, and found another subtle bug which I added a fix for.
Furthermore, improved testHybridEstimation.cpp quite a bit.