Add analytical Jacobian for EdgeSE2PointXYBearing type

[matthew-t-watson]

• Add analytical Jacobian for EdgeSE2PointXYBearing type
• Add a tolerance parameter to the evaluateJacobian helper method to allow tests to specify a different Jacobian tolerance value

[matthew-t-watson]

@RainerKuemmerle it looks like the Jacobian unit test was just failing due to a small difference between the numeric and analytical Jacobians. I think this error is actually being introduced by the numeric approach, as increasing the delta used when computing this from 1e-9 to 1e-8 at

g2o/g2o/core/base_fixed_sized_edge.hpp

Line 161 in de58780

constexpr double delta = cst(1e-9);

makes the unit test pass. I suspect this delta is too small to generate an accurate difference for certain values of the error function, likely either due to precision issues in (1/(x^2 + y^2) or atan2.

I obviously don't want to globally change the numerical difference delta size to make this pass, but given the closeness of the results and workaround I'm confident my new code is correct, so don't really want to complicate the test to work around what appears to be an existing issue. I think the best route forward is to parametrise the tolerance in

g2o/unit_test/test_helper/evaluate_jacobian.h

Line 65 in de58780

void evaluateJacobian(EdgeType& e, JacobianWorkspace& jacobianWorkspace,

and use a larger value for this specific test. Thoughts?