Phase 21.7 normalization: optimization pre-work + bench harness expansion

- Add opt-in optimizations (defaults OFF)
  - Ret purity verifier: NYASH_VERIFY_RET_PURITY=1
  - strlen FAST enhancement for const handles
  - FAST_INT gate for same-BB SSA optimization
  - length cache for string literals in llvmlite
- Expand bench harness (tools/perf/microbench.sh)
  - Add branch/call/stringchain/arraymap/chip8/kilo cases
  - Auto-calculate ratio vs C reference
  - Document in benchmarks/README.md
- Compiler health improvements
  - Unify PHI insertion to insert_phi_at_head()
  - Add NYASH_LLVM_SKIP_BUILD=1 for build reuse
- Runtime & safety enhancements
  - Clarify Rust/Hako ownership boundaries
  - Strengthen receiver localization (LocalSSA/pin/after-PHIs)
  - Stop excessive PluginInvoke→BoxCall rewrites
- Update CURRENT_TASK.md, docs, and canaries

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

docs/guides/perf/benchmarks.md

@@ -0,0 +1,67 @@
+# Benchmarking MIR Generation & AOT (Quickstart)
+
+This guide shows how to measure Hakorune's MIR emit (Stage‑B → MIR(JSON)) and AOT build (MIR(JSON) → obj/exe) without llvmlite. All commands are semantics‑preserving and keep defaults conservative (fail‑fast, O0).
+
+## Prerequisites
+
+- Build binaries once (release):
+  - `cargo build --release`
+  - `cargo build --release -p nyash-llvm-compiler` (ny-llvmc)
+  - `cargo build --release -p nyash_kernel` (NyRT static runtime)
+
+## Scripts
+
+### 1) MIR emit bench (Stage‑B → MIR(JSON))
+
+- Script: `tools/perf/bench_hakorune_emit_mir.sh`
+- Usage: `tools/perf/bench_hakorune_emit_mir.sh <input.hako> [rounds]`
+- Output CSV: `round,ms,size,sha1` (sha1 is normalized JSON digest; identical = structure equal)
+- Useful env toggles:
+  - `HAKO_USING_RESOLVER_FIRST=1` (resolver‑first)
+  - `HAKO_SELFHOST_BUILDER_FIRST=1` (selfhost builder → provider fallback)
+  - `HAKO_MIR_BUILDER_BOX=hako.mir.builder|min` (builder selector)
+  - `HAKO_SELFHOST_TRACE=1` (stderr trace)
+
+Example:
+
+```
+tools/perf/bench_hakorune_emit_mir.sh apps/examples/json_query/main.hako 5
+```
+
+### 2) MIR(JSON) → obj/exe bench（ny-llvmc / crate backend）
+
+- Script: `tools/perf/bench_ny_mir_builder.sh`
+- Usage: `tools/perf/bench_ny_mir_builder.sh <mir.json> [rounds]`
+- Output CSV: `kind,round,ms`（kind = obj|exe）
+- Useful env toggles:
+  - `NYASH_LLVM_BACKEND=crate`（既定。ny-llvmc を使う）
+  - `HAKO_LLVM_OPT_LEVEL=0|1`（既定は 0／O0）
+
+Example:
+
+```
+tools/perf/bench_ny_mir_builder.sh /path/to/out.json 3
+```
+
+### 3) MIR(JSON) 構造比較
+
+- Script: `tools/perf/compare_mir_json.sh`
+- Usage: `tools/perf/compare_mir_json.sh <a.json> <b.json>`
+- 出力: サイズと正規化 SHA1、差分（jq -S 利用時は整形差分）。
+
+## Typical Workflow
+
+1) Emit MIR(JSON)
+   - `tools/hakorune_emit_mir.sh apps/APP/main.hako out.json`
+2) Measure MIR emit time
+   - `HAKO_USING_RESOLVER_FIRST=1 tools/perf/bench_hakorune_emit_mir.sh apps/APP/main.hako 5`
+3) Measure AOT（obj/exe）
+   - `NYASH_LLVM_BACKEND=crate tools/perf/bench_ny_mir_builder.sh out.json 3`
+4) Compare MIR outputs across toggles/branches
+   - `tools/perf/compare_mir_json.sh out_before.json out_after.json`
+
+## Notes
+
+- All benches are best‑effort micro‑measurements; run multiple rounds and compare medians.
+- Keep defaults strict: resolver/selfhost togglesは明示時のみON。AOTは O0 既定（`HAKO_LLVM_OPT_LEVEL` で上げられます）。
+