fix: clean worker indexes on re-registration

[DatenpflegeNordHL]

Summary

• clean the previous worker's secondary indexes before replacing a worker registered with the same URL
• preserve the existing worker ID while rebuilding model, type, and connection indexes
• serialize registry mutations so cleanup and re-registration cannot interleave
• add regression coverage for stale and duplicate secondary-index entries

Testing

• "cargo test --lib core::worker_registry::tests::test_reregister_worker_cleans_secondary_indexes -- --exact"
• "cargo test --lib core::worker_registry::tests"
• "cargo fmt --check"
• "cargo clippy --all-targets --all-features -- -D warnings"
• "cargo test --lib"

Fixes #255

[hsliuustc0106]

Reviewed at 64caf081. Correct fix: previously, re-registering a worker by URL (e.g. after a model/type/label change) left stale entries in type_workers, connection_workers, and the model indexes. register() now purges all secondary indexes via remove_inner while preserving the WorkerId, and the new mutation_lock serializes the multi-map mutations (no lock nesting — public remove/remove_by_url route through the same inner path). The test covering label change + type change + removal is thorough. No findings.

---

From an automated daily review pass over new/updated PRs (head SHA frozen at 64caf081).