emit-portable: make streamEq self-check opt-in (S6a, #57)

[johnsoncodehk]

Summary

Slice 6a of #57. The streamEq self-check (full relex + field-by-field comparison after every edit()) was acceptance scaffolding for the windowed-relexing slices (S3–S5b) and cancelled their lexing win. It is now opt-in:

• TS createDoc(src, { validate? }), Go Doc.SetValidate, Rust Doc::set_validate — default off. With validation off, the post-edit rescan is skipped and the alignment JSON omits the streamEq key ({oldN,newN,prefix,suffix,relexed}); no fabricated streamEq: true.
• The edit-session runner protocol keeps validation ON, so the gate and all adversarial corpus scripts continue to exercise the checked path unchanged. A new edit-session-fast protocol runs with validation off.
• The gate now runs every edit scenario in both modes and asserts the five alignment fields, CST bytes, and exit codes are identical — catching flag-plumbing divergence between the paths.
• The windowed-relex logic itself (windowRelexStep, state restoration, resync) is untouched line-for-line.
• Perf reference (informational): 30k-char calc doc, 100 single-char edits, TS target — 418 ms validated vs 206 ms fast. Full reparse still dominates; tree reuse is the next milestone.

Validation

• node test/portable-targets.ts — 34 scenarios × 3 targets × 2 modes, fast ≡ validated everywhere
• node test/check.ts emit-parity — 5/5; node test/newline-mode.ts — 29/29; npx tsc --noEmit — clean; 5 corpus scripts ALL PARITY (validated path, zero changes)
• Mutation test: making the fast path keep a stale token stream fails the multi-batch fast≡validated assertions, reverted before commit
• Independent probe: multi-batch inheritance chains (javascript / templatejs / envspec) × 3 targets — five fields equal, no streamEq key in fast mode, byte-identical stdout and exit codes