remove dead clear_partition for broadcast catalog stores

[fabracht]

Summary

Follow-up to #61. With schemas and constraints now treated as broadcast state on the export side (every partition snapshot carries the full catalog), the per-partition clear_partition methods on SchemaStore and ConstraintStore are wrong-by-construction — they let a partition wipe drop catalog entries that every node is supposed to hold.

Audit finding before deleting:

• StoreManager::clear_partition (the aggregator at partition_io.rs:206-224) has zero callers anywhere in the codebase — not production, not tests. Pure dead code.
• SchemaStore::clear_partition and ConstraintStore::clear_partition are referenced only by their own unit tests (clear_partition_removes_only_target). Nothing in the migration, rebalance, or snapshot paths calls them.
• import_partition (snapshot.rs:95) does not clear before importing, so the broadcast invariant established by bundle full schema and constraint catalogs in partition snapshots #61 is intact today.

Rather than leaving a comment warning future readers not to wire these into a partition-relinquishment path, this PR removes the footgun outright:

• Deletes StoreManager::clear_partition (15-line dead aggregator).
• Deletes SchemaStore::clear_partition + its unit test.
• Deletes ConstraintStore::clear_partition + its unit test.

Per-partition clear_partition on DataStore, IndexStore, UniqueStore, FkStore, and the MQTT-side stores (sessions, retained, topics, etc.) is unchanged — those are genuinely partition-scoped.

83 deletions, 0 additions, 3 files.

Test plan

• cargo make clippy — clean (pedantic, all targets + wasm).
• cargo make format-check — clean.
• cargo make test — full workspace clean.
• Pre-commit hook (format-check + clippy) ran on the commit and passed.
• grep -rn 'clear_partition' across crates/ confirms no residual references to the removed methods.

Depends on

#61 — this PR is conceptually a follow-up; it makes sense regardless of merge order but the motivation comes from the broadcast-catalog model #61 establishes.

[fabracht]

Coordination with #61

This PR conceptually depends on #61 (broadcast-catalog snapshot export) but the two are mergeable independently — no line-level overlap in the three shared files (constraint_store.rs, schema_store.rs, partition_io.rs):

• bundle full schema and constraint catalogs in partition snapshots #61 modifies the export methods and adds tests.
• remove dead clear_partition for broadcast catalog stores #62 modifies struct doc comments, deletes clear_partition blocks + their tests, and removes the unused aggregator in partition_io.rs.

Auto-merge expected in either order.

Version + CHANGELOG

This PR now carries the mqdb-cluster 0.3.4 → 0.3.5 bump and a CHANGELOG entry covering the dead-code removal. #61 currently does not touch Cargo.toml or CHANGELOG.md.

Recommended merge order: #61 first, then this PR. The CHANGELOG entry references #61 by number and reads naturally in that order.

If #61 should carry the 0.3.5 bump instead: this PR rebases to 0.3.6 and the CHANGELOG entry becomes its own stanza. Trivial swap.