feat(stdlib): bounded search (find_in/rfind_in/count_in)

[zaidejjo]

Follow-up to #282/#283: per-line use of unbounded search is O(n) per line, O(n²) per file (measured 3.5s vs 211ms on an 8MB tree). The bound must live in the primitive.

API (byte offsets, total functions, backend-identical)

• str.find_in(s, sub, start, end) — first match in [start, end)
• str.rfind_in(s, sub, start, end) — last match in [start, end)
• str.count_in(s, sub, start, end) — non-overlapping matches in range
• Empty patterns: find/rfind return clamped start, count returns 0
• Sliceless byte windows (no boundary checks needed: valid patterns can't match mid-char, and no slicing means no panics on any input)

Layers

Rust natives (memchr) + checker sigs + C twins (memchr-skip) + headers + lowering + fixture cases + parity (existing entries) + docs.

Verify

• e2e (VM) ✓, dual-engine parity (VM+AOT) ✓, stdlib lib ✓, clippy + fmt clean