Let a self-referential let-bound closure resolve itself as an upvalue

[revarbat]

Summary

• Follow-up to Compile captured closures' own bodies, not just their creation #53. reduce()/accumulate()/while()'s own idiom — let(a = function(...) ... a(...) ...) a(...), a closure that calls itself by name — was excluded from compiled invocation entirely (correct, but permanently interpreted) to avoid an O(n²) blowup, because the self-reference never resolved as a genuine upvalue: LetOp's declareLocal for a ran only after compiling a's own RHS.
• A direct let(name = function(...) ...) binding's own slot is now declared before compiling that RHS (letrec-style), so a self-reference resolves as Op::LoadUpvalue like any other capture. Op::MakeClosure's runtime handler can't read this one out of slots the way it does every other capture — the closure being built doesn't exist yet at the moment its own captures are normally snapshotted — so it's deferred and patched into the closure's own capturedLet immediately after construction instead.
• Confirmed via the CLI: a 32,000-element reduce() that previously took 7+ seconds now runs in ~0.05s, correct at every scale tested up to 500,000 elements.
• Deliberately scoped to direct self-reference — a closure further wrapped in a ternary, or mutual recursion between two sibling let-bound closures (confirmed already broken before this change, in the released v0.8.0), still falls back to interpreted via the existing containsLoadFree guard.

Test plan

• Full test suite (671 tests, up from 670) passes locally.
• Renamed/updated SelfReferentialRecursiveClosureNowResolvesAsUpvalueAndRunsCompiled (was ...StillRunsInterpretedAndCorrectly) — confirms both correctness and that it now runs compiled (11 stops vs. 30 before, diffed against the pre-fix build).
• New PlainLetSelfReferenceStillSeesTheOuterBindingNotItself — confirms the letrec pre-declare is scoped to function-literal RHS only; let(x = x + 1) still correctly sees the outer x.
• Manually verified via the CLI: count_to/f_1arg (from Compile captured closures' own bodies, not just their creation #53) unaffected; ternary-wrapped self-reference and fnliterals.scad's real accumulate()/while() all still correct.

🤖 Generated with Claude Code