[3.x] Tight shadow culling - increase epsilon to prevent flickering

[lawnjelly]

Near colinear triangles were still causing inaccuracy in culling planes, so the threshold for colinearity is bumped up.

Fixes #91976
3.x version of #92078

Notes

• In the MRP in the issue, the spotlight was placed on the exact boundary of the top and bottom frustum planes, this was causing sometimes frustum places to be added and sometimes not (due to float error). This in itself was not a problem.
• The triangles formed between the light position and the edges however were proving too close to colinear, which as before when converted to a Plane caused inaccurate normal and d values.
• Originally I set the epsilon for colinearity deliberately very low, to see what we could get away with. It turns out it needs to be higher so I've gone for quite a conservative value here.

[lawnjelly]

Thanks!