Update singularity check for KB camera model

[nickcharron]

In this paper: https://arxiv.org/pdf/1807.08957.pdf

They describe that KB has a singularity at z = 0. In our InProjectionDomain function, we only check that z != 0

See check here:

libbeam/beam_calibration/src/KannalaBrandt.cpp

Line 143 in 4384b41

bool KannalaBrandt::InProjectionDomain(const Eigen::Vector3d& point) {

Since it's a singularity, I think we should exclude points that are 'near' z = 0, not exact, as points approaching zero will probably give errors as well. Especially with machine rounding errors

@jakemclaughlin6 @adthoms (should be a quick fix if either of you have a minute)