refactor: reorganize bytecode tests with file-based snapshots

[zharinov]

Summary

• Reorganize bytecode emission tests from inline to file-based snapshots
• Delete dump_tests.rs, consolidate all tests into codegen_tests.rs
• Comprehensive coverage organized by language feature (51 tests)
• Improve dump format readability

Dump format changes

Section names simplified:

• [header] → [flags]
• [types.defs] → [type_defs]
• [types.members] → [type_members]
• [types.names] → [type_names]
• [entry] → [entrypoints]
• [code] → [transitions]

Type definitions use cleaner syntax:

- T03 = Struct(M0, 1)  ; { name }
+ T03 = Struct  M0[1]  ; { name }

Members/names use arrow syntax:

- M0 = (S01, T02)  ; name: str
+ M0: S01 → T02  ; name: str

Test organization

51 tests organized by language feature:

1. Nodes (named, anonymous, wildcard, error, missing)
2. Captures (basic, multiple, nested, typed)
3. Fields (single, multiple, negated)
4. Quantifiers (optional, star, plus, non-greedy)
5. Sequences (basic, nested, with captures)
6. Alternations (unlabeled, labeled, in quantifiers)
7. Anchors (between siblings, first/last child)
8. Named expressions (definitions, references)
9. Recursion (simple, structured result)
10. Optionals (first child, null injection)

Test plan

• All 51 tests pass