[3.x] Float literals - fix math classes to allow 32 bit calculations

[lawnjelly]

Converts float literals from double format (e.g. 0.0) to float format (e.g. 0.0f) or integer where appropriate for 32 bit calculations, and cast to (real_t) or (float) as appropriate.

This ensures that appropriate calculations will be done at 32 bits when real_t is compiled as float, rather than promoted to 64 bits.

Combined version of #57703 and #58488 for 3.x

Notes

• I opted to combine these into one PR for 3.x, just saves me going through the same code twice.
• In contrast to the 4.x version this also covers the BVH.

[aaronfranke]

As with the other PR, the only notable behavior changes are all things that we don't care much about precision of, such as comparisons with 0.99 or 0.0001.

Unlike the other PR, for 3.x we can assume real_t will always be float, so adding f suffixes is safer here since calculations should be 32-bit anyway.

[akien-mga]

Thanks!