TypedPython: numpy float floor_divide/remainder warnings differ from numpy on Linux (glibc) #292

Since #264, test_buffer_numpy p_fdiv, p_mod, i_fdiv, i_mod fail on ubuntu CI (numpy 2.5.4, same as macOS where they pass). p_mod/i_mod (0.0 % 0.0): numpy on Linux reports 'invalid value encountered in scalar remainder', compiled code clears the flags after fmod by zero (tp_np_divmod, tp_runtime.h around line 479). p_fdiv/i_fdiv (0.0 // nan): compiled code reports invalid, numpy does not (likely a NaN comparison raising FE_INVALID under gcc/x86). Compiled code must report exactly what numpy reports on the same platform.

Done when: test_buffer_numpy passes on ubuntu and macOS CI.