nx_camera_q14.nx

nx_camera_q14.nx -- camera + projection + 4x4 matrix math in Q14. /voxels runtime primitive per nxc2/docs/NISHI_GAME_ENGINE_ROADMAP.md. The math layer that powers view, projection, and per-vertex transform for the software rasterizer (queued). Flat-buffer API: every matrix is a 16-element i64 array (row-major, Q14 entries). Every vector is a 4-element i64 array. No struct usage by design -- this primitive stays WAT-clean for browser delivery today. Once self-host closes OP_GEP-on-WAT, a companion nx_camera_struct.nx will wrap these flat-buffer helpers behind a natural CameraState struct. Both APIs ship side-by-side after that. Q14 convention (matches nx_matrix.nx + nx_sound.nx): 1.0 = 16384. Compose freely with existing primitives that use Q14. Trig: 19-entry sin table at 5-degree resolution covering 0..90. Symmetry folds extend to 0..360. Linear interpolation between table entries yields ~0.5 degree max error -- acceptable for general camera control where the eye cannot distinguish under 1 degree. FULL CAPABILITY (per feedback-maximum-capability-no-simplification cardinal, 2026-05-16): higher-resolution variants use Bhaskara I's sin approximation (closed-form rational, accurate to ~0.0017 peak absolute error = ~28 Q14 units, no table required). Variants: nx_camera_sin_q14_centideg(d100): input in centi-degrees, range [-36000, 36000] -- 0.01-degree precision nx_camera_sin_q14_millideg(d1000): input in milli-degrees, range [-360000, 360000] -- 0.001-degree precision Both variants use the same Bhaskara closed-form, scaled to the integer input unit. The 5-degree table remains the legacy API for callers that don't need sub-degree precision. Loss audit: - Q14 multiplications lose 14 fractional bits per multiply. For