Portable targets: document session + edit API (S1, full-reparse fallback)

[johnsoncodehk]

Summary

First slice of #57: all three portable targets (TS/Go/Rust) gain a document session + edit API. The API shape is now fixed; later slices swap the internals for true incremental reuse without touching the interface.

• TS: createDoc(src) → { text(), root(), edit(edits) } with Edit = { start, end, text }
• Go: Doc struct + NewDoc / Text / Root / Edit
• Rust: Doc holds the owned String; edit splices text, parse() borrows on demand (sidesteps the Parser<'a> self-reference)
• Edit semantics (uniform): each (start, end, text) replaces [start, end); edits apply sequentially; offsets in each target's native CST offset units
• Internals delegate to the existing tokenize/parse (full re-parse — no incremental reuse in this slice, by design); existing lexer/parser function bodies are append-only unchanged
• Runner gains an edit-session mode ({init, batches} JSON on stdin, same CST JSON output), so the gate can adjudicate sessions

Validation

• test/portable-targets.ts: 8 scripted edit scenarios (insert/delete/replace/multi-batch, including same-batch offset chaining) × 3 targets — all edit-session ≡ fresh parse of final text byte-identical + oracle parity; all existing accept/reject counts unchanged
• node test/check.ts emit-parity — 5/5
• npx tsc --noEmit — clean
• Planner adversarial probe: 4 extra scenarios outside the gate (reject-final, empty-final, append-at-end, reject-intermediate-state) × 3 targets — 12/12 parity

Part of #57 (S1 of the slicing plan; S2+ — windowed relex, arena persistence, adoption — tracked separately).