Distinguish composite types of user types from built-in composites

[cberner]

Problem

A composite type (Option, Vec, tuple, or array) built its TypeName from only the inner type's name string, discarding the inner type's user-defined-vs-built-in classification. So Option<u32> (built-in) and Option<MyType>, where MyType is a user type named "u32", produced byte-identical type names.

A table created with one could then be opened as the other with no error, silently reading the stored bytes under the wrong encoding and yielding wrong results or unreachable keys. The classification byte exists precisely to keep a user type named "u32" from colliding with the built-in u32, but that protection was lost one nesting level down.

Fix

Bubble the classification up: a composite that wraps a user-defined type is now itself classified user-defined, keeping the same name string. Composites of only built-in types are byte-for-byte unchanged, so existing databases open unchanged.

To keep tables created by older versions readable, TypeName carries an in-memory-only record of the classification it was bubbled from (never serialized), and check_match accepts that prior spelling in addition to the current one:

• For fixed-width composites the prior spelling is Internal, which also matches redb 2.6.
• For variable-width tuples it is Internal2 only, so redb 2.6's incompatible variable-tuple encoding is still correctly rejected rather than silently misread.
• A flat user-defined type records no such classification, so this never lets a user type named "u32" alias the built-in u32 at the top level.

The Value/Key public traits are untouched — the whole mechanism lives in TypeName plus the composite type_name() impls (which already had to change for the fix).

Known limitation

An old Option<user-type> table is byte-identical on disk to a built-in Option<u32> table, so a name-colliding user type can still open a built-in table in the reverse direction. This is inherent to preserving backwards compatibility and only reachable if a user names their type to collide with a built-in (which the docs already advise against). It is strictly better than before, where the alias happened unconditionally in both directions.

Tests

• Crate-internal unit tests in src/types.rs: built-in composites keep their exact classification (unchanged on-disk bytes); a composite of a name-colliding user type no longer equals its built-in counterpart; matches_legacy accepts the right Internal/Internal2 prior spellings; and a flat user type never legacy-matches an Internal name.
• Cross-version tests in tests/backward_compatibility.rs using the real redb 2.6.0 dev-dependency: an Option<user-type> table written by redb 2.6 still opens and reads back correctly, and an Option<user-type> table can no longer be opened as the built-in Option<u32> (returns TableError::TableTypeMismatch).

cargo fmt --check, cargo clippy --all-targets --all-features, and the full cargo test --all-features suite all pass.

🤖 Generated with Claude Code

---

Generated by Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 89.51613% with 13 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.22%. Comparing base (b6b56aa) to head (1eddc21).

Files with missing lines	Patch %	Lines
src/types.rs	87.96%	13 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #1298    +/-   ##
========================================
  Coverage   90.21%   90.22%            
========================================
  Files          36       36            
  Lines       16330    16447   +117     
========================================
+ Hits        14732    14839   +107     
- Misses       1598     1608    +10     

Files with missing lines	Coverage Δ
src/complex_types.rs	87.50% <100.00%> (+0.40%)	⬆️
src/tree_store/table_tree_base.rs	89.38% <100.00%> (+0.21%)	⬆️
src/tuple_types.rs	91.74% <100.00%> (+2.95%)	⬆️
src/types.rs	96.22% <87.96%> (-2.84%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.