PnP update

[marip8]

Per investigation in #57, this PR provides a workaround for using Eigen objects without local paramterization. It also adds a function for calculating the "full" covariance matrix between two parameter blocks (i.e. variance with the parameter block itself and covariance with the other parameter block)

[marip8]

@schornakj can you review? This is ready to merge but has a few changes that might overlap with #62

[schornakj]

@schornakj can you review? This is ready to merge but has a few changes that might overlap with #62

I'll rebase onto your changes after we merge. #62 adds a function that's similar (but uses different underlying Ceres functions) to the "full" covariance function you added here so you might be interested in taking a look and seeing which approach you prefer.

[marip8]

@schornakj I addressed your comments in the latest commits. I ended up running the randomly initialized optimizations 30 times and setting expectations that the mean error + 3 standard deviations should be less than a (still somewhat arbitrary) threshold. I think this is at least a step in the right direction for setting reasonable expectations for unit tests with randomness

[schornakj]

@schornakj I addressed your comments in the latest commits. I ended up running the randomly initialized optimizations 30 times and setting expectations that the mean error + 3 standard deviations should be less than a (still somewhat arbitrary) threshold. I think this is at least a step in the right direction for setting reasonable expectations for unit tests with randomness

That's reasonable. Looks good to me!