particle: contiguous vertex writes, skip normals, stride logging

- Detected interleaved 24-byte vertex layout: xyz(12)+color(4)+uv(8).
  Fast path writes 6 sequential u32s instead of scattered stores.
- Normal stride=0 (shared global) — write once in writeback, not 4x.
- Added stride_info export + logging for vertex layout analysis.
- A/B: ~30% peak reduction, baseline also improved due to less overhead.

src/transform44/transform44.zig

@@ -25,6 +25,8 @@ extern fn transformImpl_SSE(u32, u32, u32, u32, u32) callconv(.c) void;
 extern fn calcColorValues_SSE(u32, u32, u32, u32, u32, u32, u32) callconv(.{ .x86_thiscall = .{} }) void;
 extern fn renderParticleSprites_SSE(u32, u32, u32) callconv(.{ .x86_thiscall = .{} }) u32;
 extern fn resetParticleCache() void;
+extern var stride_info: [8]u32; // exported from particle_sse.zig
+var stride_dumped: bool = false;
 
 /// Thiscall wrapper for the SSE implementation. Lives here (baseline SSE2 unit)
 /// so LLVM can't inline transformImpl_SSE's alignment into the thiscall frame.
@@ -1458,6 +1460,17 @@ fn dumpStats() void {
         });
     }
 
+    // Dump particle VB stride info (once)
+    if (stride_info[0] != 0 and !stride_dumped) {
+        stride_dumped = true;
+        log.fmt("  vb_strides: pos={d} norm={d} color={d} tc={d}\n", .{
+            stride_info[0], stride_info[1], stride_info[2], stride_info[3],
+        });
+        log.fmt("  vb_bases: pos=0x{x} norm=0x{x} color=0x{x} tc=0x{x}\n", .{
+            stride_info[4], stride_info[5], stride_info[6], stride_info[7],
+        });
+    }
+
     // Flip A/B mode
     ab_use_custom = !ab_use_custom;
     diag_cmp_count = 0;