emit-portable: windowed relexing for newline-mode lexers (S4, #57)

[johnsoncodehk]

Summary

Slice 4 of #57 (portable incremental edit). Extends S3's windowed relexing (#64) from stateless lexers to newline-mode lexers (e.g. envspec):

• Verified state analysis: of the newline lexer's five cross-token state variables (pos, pendingNl, lineStart, emittedContent, flowDepth), the first four hold constant values at every token-emission point — the only one that varies is flowDepth. So each token's AlignMeta now records the flow depth after its emission (fd), and a windowed scan can restart from any token boundary with state {pos: end(anchor), pendingNl: false, lineStart: false, emittedContent: true, flowDepth: fd(anchor)}.
• Resync additionally requires the old token's recorded fd to equal the live scan depth — an insertion of ( that shifts downstream flow depth correctly refuses to adopt the old suffix (gate scenario dedicated to this). Zero-width NEWLINE tokens sharing an offset with their successor are disambiguated by kind during resync lookup.
• The streamEq self-check compares fd too, so an adopted-but-wrong depth cannot survive as a latent state bomb for the next edit.
• Drive-by parity fix: the portable newline lexer used to set pendingNl = true on LF/CR inside flow suspension; the interpreter does not (non-indent newline mode). Removed — direct token-stream comparison against the oracle lexer over flow/newline-heavy inputs is now field-identical. (Latent in S3: CST output doesn't carry nl, and the old envspec edit scenario forced full relex.)
• regexCtx/template grammars (javascript) keep the full-relex fallback; their lexer codegen is untouched.
• Gate: envspec grows 1 → 8 edit scenarios (flow-interior edit, depth-shift insert, comment edit, blank-line fold, large mid/tail with relexed ≤ 10, two-batch depth-inheritance chain); 21 scenarios × 3 targets total.

Validation

• node test/portable-targets.ts — calc 8/8 + javascript 5/5 + envspec 8/8, × 3 targets; all streamEq true
• node test/check.ts emit-parity — 5/5; node test/newline-mode.ts — 29/29; npx tsc --noEmit — clean
• Mutation test: disabling the fd resync check makes the depth-shift scenario fail (streamEq:false + align mismatch), reverted before commit
• Independent adversarial probe (6 attacks × 3 targets): delete flow-close/open, nested depth-2 interior edit, resync at zero-width NEWLINE neighborhood, three-batch depth-inheritance chain, edit between two flow regions — all parity with oracle-derived expectations
• Direct oracle token-stream parity over 7 flow/newline-heavy inputs (field-by-field)