wasm-gc: tuple elements keep generic payload types; Char.toCode returns Unicode scalars

[jasisz]

Fixes the two wasm-gc residuals documented in the 0.25.0 changelog, plus a type-checker bug the repro matrix surfaced.

What was broken

• Tuple<Option<Int>, Int> with an Option.None branch failed to compile on wasm-gc with cannot lower type `Option<T>` to a wasm representation — the None literal nested in a tuple element never inherited the concrete payload type. Same for tuples holding Result (verify expectations like (Result.Ok(2), Result.Ok(0))). VM ran both fine.
• Char.toCode on wasm-gc returned the first UTF-8 byte instead of the Unicode scalar: 'ż' gave 197 instead of 380, '😀' gave 240 instead of 128512. ASCII masked the bug.
• Bonus from the discriminator probes: (Result.Ok(n), n) against a declared Tuple<Result<Int,String>, Int> was a false type error even on the VM.

The fix

• Tuple literals in driven positions (declared return, annotated binding, verify expectation) adopt the expected Tuple<...> element-wise — the same bidirectional context list elements already receive. Side effect, now documented in the changelog: an element that doesn't fit the annotation is rejected at check time instead of failing at run time.
• Char.toCode lowers through a demand-driven helper that decodes the first UTF-8 scalar (2/3/4-byte lead masks per spec). Modules that never call it are byte-identical to before.
• The expected-type path deliberately excludes (...)! products, so their "element must be a function call" validation still fires — pinned by new type-checker tests (this was caught in adversarial review of the first version of this change).

Verification

• Repro matrix of 19 probes (16 canonical + 3 discriminators), every probe VM-green; failing rows now pass on wasm-gc; the Some-only controls confirm no behavior change where things already worked.
• Adversarial review in a fresh worktree: 13 extra probes (4-byte emoji, multi-byte char equality, empty strings, deep tuple nesting, two-module program sharing the same tuple instantiation, legal (f(), g())!), byte-reproducibility cmp 3×, zero drift on untouched modules (order_total.wasm byte-identical), test honesty revert-checked (new tests fail without the source fix).
• Independent second-opinion review (different model family) cross-checked the UTF-8 mask math and enumerated checker entry points; its remaining finding — hand-written == between such tuples outside verify — stays a documented residual rather than a speculative fix.
• cargo test --workspace --features wasm,wasip2 (80 suites), fmt, clippy: green. wasip2 smoke compile with Char.toCode: green.

🤖 Generated with Claude Code