Resolve mutual recursion between sibling let-bound closures

[revarbat]

Summary

• Extends the letrec pre-declare from Let a self-referential let-bound closure resolve itself as an upvalue #54 (self-reference) to cover two-or-more sibling let-bound closures that reference each other by name — fnliterals.scad-style isEven()/isOdd(), each calling the other. All function-literal-RHS assignments in a LetOp now get their own slot pre-declared up front (not just each one's own, right before compiling it), so a forward reference to a not-yet-compiled sibling also resolves as a genuine upvalue.
• Unlike self-reference, a forward reference has nothing to read or self-patch into at Op::MakeClosure time — the sibling it names doesn't exist at all yet. Such captures are left out of the closure's own capture list entirely at compile time and deferred to a new Op::PatchClosureCapture instruction, emitted right after the referenced sibling's own StoreLocal actually runs — mutating the earlier closure's already-real capturedTrail in place.
• Verified 2-way and 3-way (f1->f2->f3->f1) cycles both resolve correctly, including a case mixing a genuine outer capture with a forward-sibling reference in the same closure. Both scale linearly (100,000-deep isEven/isOdd chain in ~0.07s) — no repeat of the capturedLet-chaining regression the self-reference fix needed a guard against, since a resolved forward reference is a real capture snapshotted once, not re-derived every call.

Test plan

• Full test suite (673 tests, up from 671) passes locally.
• New MutualRecursionBetweenSiblingLetBoundClosuresResolvesCorrectly and ThreeWaySiblingMutualRecursionResolvesCorrectly.
• Manually verified via the CLI: 2-way/3-way cycles, scale (10K/100K), a mixed genuine-capture + forward-reference case, and all previously-fixed patterns (count_to, f_1arg, reduce() at scale, let(x=x+1) shadowing, ternary-wrapped self-reference, accumulate()/while()) unaffected.

🤖 Generated with Claude Code