…rence (#878)

A generic whose type argument must be inferred from a user-fn call's return
type in argument position monomorphized to the Bool phantom-var default: on a
check-green program `vera run` emitted a call to a `$Bool` clone Pass 1.5 never
produced, so the caller (often `main`) was skipped and dropped from the exports
("No exported functions to call"), or a private helper left a dangling `call`
that failed WAT validation.

Root cause is the lossy WAT collapse `i32 -> "Bool"` (a Decimal handle, an ADT,
and an Option/Result pointer are all i32 — the same value as the phantom
default) applied to user-fn returns on both the discovery and the body-emission
sides:

- Discovery (vera/codegen/monomorphize.py): `_build_mono_context` seeded the
  shared `fn_ret_types` from WAT signatures, so a generic bound solely by a
  user-fn return was discovered as `$Bool` while the #732 verifier discovered
  the precise type — a differential desync masked only by identity clone
  bodies. Now seeds from each fn's declared return TypeExpr, exactly as the
  verifier does, with the WAT collapse kept only as a fallback.
- Body-emission (vera/wasm/inference.py): `_get_arg_type_info_wasm` had no
  FnCall branch, so a parameterized return in `Option<VeraT>` position bound
  nothing; and a user-fn call in bare `@VeraT` position collapsed i32 to Bool.
  It now mirrors discovery's FnCall branch (shared _BUILTIN_PARAMETERIZED_RETURNS
  plus a non-generic-user-fn parameterized case) and consults the declared
  return TypeExpr for the ambiguous i32 case. GENERIC calls are excluded from
  the user-fn branch (their declared return is over the callee's own type vars,
  not concrete types) so they fall through to generic-return resolution exactly
  as discovery does — keeping the two consultors in lockstep.

Test values are deliberately Decimal returns (`0.3333...`) that cannot coincide
with the Bool default (1/0), the exact Bool-coincidence trap CLAUDE.md warns
about. New run-level conformance program ch09_generic_infer_user_fn_return (suite
123) plus three regression tests (crash-via-IO, WAT-mangle discriminator,
discovery-vs-verifier differential); each change is independently mutation-
validated. Does not weaken #888's cross-module/closure discovery.

Co-Authored-By: Claude <[email protected]>