feat(stdlib): positional str search (find/rfind/starts_with_at/ends_with_at/trim_span)

[zaidejjo]

Byte offsets throughout: O(1) per call, O(n) streaming total, backend-identical on every input (matches Rust str::find byte semantics; len()/slicing stay char-oriented by design, documented).

Layers (all five, in lockstep)

• Rust natives (str_mod) + registry (natives/mod.rs, std + short spellings)
• Checker sigs (funcs.rs) + registry count/self-test updates
• C runtime twins (strings.c/.h) + AOT lowering map
• e2e fixture str_find_test.zz + registration + dual-engine parity entry
• docs/stdlib.md (signatures, offset contract, examples)

Notable decisions

• Byte (not char) offsets: char conversion is O(from) per call, which makes streaming loops O(n²) (measured 16.7s vs 21ms on 322KB before the switch; 13ms after). Char slicing composition gap documented.
• trim_span strips Unicode White_Space explicitly on both backends (never the host trim — sidesteps the native/VM trim split).
• Non-boundary positions snap (ceil find, floor rfind/tail); empty patterns return clamped positions, never match in _at.

Verify

• e2e (VM) ✓, dual-engine parity (VM+AOT) ✓, zz_stdlib 191/191 ✓, clippy + fmt clean
• Streaming find-loop line scan beats split (13ms vs 21ms x5 on 322KB) with zero substrings allocated