[BR] Fix integer overflow in physx::writeCompressedContact causing sim corruption

[Zeblote]

If a contact buffer has exactly 256 contacts in it, the endIndex on its last StrideBatch will overflow from 256 to 0, causing it to fail to copy the last batch of contacts but still setting the output count to include it, sending uninitialized garbage data into the solver, which kills the simulation very quickly.

Example of 256 input contacts to the function

Corresponding grouped output contacts, note the last group is garbage

The root cause is this u8 overflowing back to 0

We can fix it easily by using PxU16 instead so it can hold the proper count

cc @PierreTerdiman