Silent BytecodeTypeId(0) fallback in type resolution swallows a must-hold invariant

[zharinov]

Problem

Type resolution silently falls back to BytecodeTypeId(0) instead of failing loudly. This breaks the project's invariant discipline: a swallowed must-hold fact should be made loud.

After a successful analyze, every referenced type must resolve. A miss is our bug. It should panic, not silently alias to T0 and emit a wrong-typed table.

• emit/type_table.rs:127 uses .unwrap_or(BytecodeTypeId(0)) when collecting a TypeName's id.
• emit/type_table.rs:271-272 resolve_type returns Ok(BytecodeTypeId(0)) with a "should not happen for well-formed types" comment.

Both spots are inside the trust boundary (post-analyze). CLAUDE.md's "Invariant Discipline" says state invariants loudly here (panic!/.expect()/invariants.rs), not swallow them with unwrap_or/.ok().

Cause

emit/type_table.rs:127 and emit/type_table.rs:271-272 both turn an unresolved-type miss into a T0 value instead of a loud failure.

Fix

Replace both silent T0 fallbacks with a loud failure (.expect(...) / panic! / an invariants.rs ensure_*) that names the unresolved type.

The fallback was never meant to fire for well-formed queries:

• Valid input behaves the same.
• A silent miscompile becomes a caught bug.

Acceptance

• Resolving an unresolvable type panics with a descriptive message in debug/test builds instead of emitting T0.
• No behavioral change for well-formed queries.

[zharinov]

Fixed by #479 (10fd1836). Both silent BytecodeTypeId(0) fallbacks are gone:

• emit/type_table.rs:127 now .expect("def result type must be mapped") (was .unwrap_or(BytecodeTypeId(0))).
• resolve_type (emit/type_table.rs:263) now .expect("resolved type must be mapped") (was Ok(BytecodeTypeId(0)) with the "should not happen for well-formed types" comment).

An unresolvable type now panics with a descriptive message instead of aliasing to T0, and there's no behavioral change for well-formed queries — matching the acceptance criteria. A repo-wide grep for unwrap_or(BytecodeTypeId(0)) / Ok(BytecodeTypeId(0)) returns nothing.