chore(#191): gitignore profiling artifacts, shrink verifier stack, document profiling

- .gitignore: perf.data*, sizecheck, and the built go-embed example binary
- chunk_verifier: func_ips as [4096]u32 (16KB stack instead of 32KB on
  64-bit; ips are bounded by MaxCode = 1 MiB)
- CONTRIBUTING: profiling recipe — --call-graph dwarf is required for
  trustworthy call graphs on the ReleaseFast binary (frame-pointer unwinding
  misattributes inlined frames), plus pointers to -Dperf counters,
  gc_stats_ext, and the bench baselines

Author: ganehag

.gitignore

@@ -22,3 +22,8 @@ gengo_acl.so
 site/
 examples/bundle-c/quiz_grader
 examples/bundle-c/scoring.zip
+
+# Profiling and local analysis artifacts
+perf.data*
+sizecheck
+examples/go-embed/go-embed

CONTRIBUTING.md

@@ -44,6 +44,26 @@ make parity
 
 `parity` checks that the native and WASM backends produce the same output. Run it when touching the VM or compiler.
 
+## Profiling
+
+Build the timing binary and profile with DWARF call graphs:
+
+```bash
+zig build -Dpreset=1m cli-fast
+perf record --call-graph dwarf -F 999 -- ./zig-out/bin/gengo-fast tests/bench/007_dispatch_loop.gengo
+perf report
+```
+
+Do not trust call-graph percentages from plain `perf record -g` on the
+ReleaseFast binary: frame-pointer unwinding misattributes inlined frames
+(e.g. allocation helpers showing up hot inside functions that never
+allocate). `--call-graph dwarf` resolves inline frames correctly.
+
+For per-opcode counters and allocation stats, build with `-Dperf=true` and
+check `std.core.gc_stats_ext()` from scripts (`alloc_object_calls`,
+`gc_runs`, `gc_time_ns`). Timing baselines for the benchmark set live in
+`tests/bench/time_baseline.txt`.
+
 ## Docs
 
 The public docs site is sourced from `docs/` and built by Gengoscript itself

src/lang/chunk_verifier.zig

@@ -259,14 +259,16 @@ pub fn verify(state: anytype) !void {
 
     {
         var func_body_count: usize = 0;
-        var func_ips: [4096]usize = undefined;
+        // u32 halves the stack footprint (16KB vs 32KB on 64-bit) — ips are
+        // bounded by MaxCode (1 MiB), well inside u32.
+        var func_ips: [4096]u32 = undefined;
         {
             for (state.consts[0..state.const_count]) |cv| {
                 if (cv == .object) {
                     switch (cv.object.*) {
                         .function => |f| {
                             if (f.ip < state.code_len and Bits.has(starts, f.ip)) {
-                                func_ips[func_body_count] = f.ip;
+                                func_ips[func_body_count] = @intCast(f.ip);
                                 func_body_count += 1;
                             }
                         },