perf: decode rewrite, encode zeros optimization, iai-callgrind benchmarks (0.3.1)

Decode rewritten with shrinking-slice pattern (split_first/split_at)
instead of index arithmetic. Decode zeros 14% faster, decode nonzero
flips from 5% behind corncobs to 2% ahead, decode mixed closes to tied.

Encode zeros fast path moved before split_at — now tied with cobs
(was 6% behind). Code size reduced to 821B (was 869B).

Replaced criterion with iai-callgrind for deterministic instruction-count
benchmarks via Valgrind. Zero run-to-run variance. Expanded to 162
benchmarks across 9 sizes.

ucobs now wins or ties 16 of 18 benchmarks vs corncobs and cobs.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -1,5 +1,28 @@
 # Changelog
 
+## 0.3.1 — 2026-03-31
+
+### Performance
+
+- **Decode: rewritten with shrinking-slice pattern** — decode zeros is 14%
+  faster (25,018 vs 29,114 instructions at 4 KB), decode nonzero flipped from
+  5% behind corncobs to 2% ahead, decode mixed closed from 8% behind to tied.
+  ucobs now wins or ties 16 of 18 benchmarks.
+- **Encode zeros: 6% faster** — zero fast path moved before `split_at` calls,
+  eliminating per-byte overhead. Now tied with cobs (previously 6% behind).
+- **Code size reduced** — 821 B total (was 869 B). Encode 429 B, decode 392 B.
+
+### Changed
+
+- **Replaced criterion with iai-callgrind** — benchmarks now measure
+  deterministic instruction counts via Valgrind instead of wall-clock time.
+  Zero run-to-run variance. Requires `valgrind` installed.
+- **Expanded benchmark coverage** — 9 payload sizes (0–4096 B) × 3 patterns
+  × 3 crates × encode/decode = 162 benchmarks, all running in ~2 minutes.
+- **Code size measurement** added to `just bench` via `just bench-size`.
+- **Mutation testing** added via `cargo-mutants` (`just test-mutants`).
+- Added `mise.toml` for dev tool management.
+
 ## 0.3.0 — 2026-03-31
 
 ### Performance