perf: move luaalloc into weirdperformance, wire clip SSE hook (1.9x)

- luaalloc moved from standalone module to weirdperformance sub-module
- ClipPolygonToSinglePlane SSE hook wired up (410->206 cyc, 2.55% CPU)
- Eliminated calling convention wrappers: clip_sse (fastcall),
  bone_sse (thiscall) match hook conventions directly

src/weirdperformance/bone_sse.zig

@@ -1101,7 +1101,7 @@ fn calcScaledInverse(this_mat: u32, out: u32, scale: f32) void {
 //         mat3(offset_vec3*), mat4(scale_float_bits)
 // =============================================================================
 
-pub fn transformImpl_SSE(this: u32, mat1: u32, mat2: u32, mat3: u32, mat4: u32) callconv(.c) void {
+pub fn transformImpl_SSE(this: u32, mat1: u32, mat2: u32, mat3: u32, mat4: u32) callconv(.{ .x86_thiscall = .{} }) void {
 
     @setEvalBranchQuota(50000);
     // =========================================================================