Branchless atan2_int16

[atan2_jimshima_test.c](https://github.com/user-attachments/files/26434162/atan2_jimshima_test.c) This is a branch less version of the code. On my desktop the difference is minimal. But maybe on rpi and other platforms it might be better. Can anyone run the test code? (Generated by Qwen3.6 Plus Preview). ``` /dev/shm/atan2_jimshima_test --speed Jim Shima atan2_int16: speed benchmark ======================================= Iterations: 1000000000 Input range: [-32512, 32768] Method Time (s) ns/call Relative ──────────── ────────── ───────── ──────── old (guards) 4.808 4.8 1.00x new (branchless) 4.793 4.8 1.00x ```