perf: fix deferred divide precision, threadlocal TLS, MSVC target

Removed deferred divide from both ray-tri functions -- the det-scaled
epsilon comparisons lose precision for near-parallel rays, potentially
accepting triangles the original rejects. Downstream SetupBoxFrustum
then writes to wrong globals near 0xCE6738 (SGroupPtr), causing
ERROR #124 on exit.

inflate_hook: replaced manual FS:0x24 thread ID pool with Zig native
threadlocal. Fixes potential crash on Wine where FS segment layout
may differ.

Build: switched from GNU to MSVC ABI (4-5K smaller per DLL, no
.eh_frame unwind tables). Added setjmp.h stub for libdeflate MSVC
build. Added noperf build variant (zig build noperf).

src/performance/weirdperformance.zig

@@ -333,8 +333,8 @@ pub fn installHooks() void {
     // if (findguid_hook.attach(0x464890, &findguidDetour) == .ok) installed += 1;
     // if (obj_delete_hook.attach(0x464920, &objDeleteDetour) == .ok) installed += 1;
 
-    // Flush GUID cache before object manager teardown
-    if (destroy_objmgr_hook.attach(0x467700, &destroyObjMgrDetour) == .ok) installed += 1;
+    // GUID cache disabled for now
+    // if (destroy_objmgr_hook.attach(0x467700, &destroyObjMgrDetour) == .ok) installed += 1;
 
     // Per-frame cache reset
     if (world_update_hook.attach(0x482EA0, &worldUpdateDetour) == .ok) installed += 1;