refactor(common): decouple PhysicalTable and ResolvedTable from Dataset

[LNSD]

Remove Arc dependency from catalog types, storing only the
essential fields (HashReference, start_block, table name, network).

• Replace ResolvedTable.dataset: Arc<Dataset> with individual fields
• Remove Dataset::resolved_tables() method
• Update PhysicalTable constructors to take Table instead of ResolvedTable
• Store dataset_reference: HashReference directly in PhysicalTable
• Update callers in dump, admin-api, and tests

[claude]

Review Summary

This PR successfully decouples PhysicalTable and ResolvedTable from Arc<Dataset>, replacing the direct dataset reference with individual fields (HashReference, start_block, table name, network). This is a well-structured refactoring that reduces coupling and makes the data flow more explicit.

Highlights

Positive changes:

• Clean separation of concerns - PhysicalTable and ResolvedTable now store only the data they need
• Removal of Dataset::resolved_tables() method simplifies the API
• Good use of BTreeSet for deduplicating dataset references in streaming_query.rs

Key items to address

1. Breaking API change (restore.rs): The url field renamed to path in RestoredTableInfo is a breaking change for HTTP clients. Please confirm this is intentional.

2. Performance consideration (physical.rs): catalog_schema() and table_ref() now return owned types instead of references, causing allocations on each call. Consider caching if these are hot paths.

3. Data duplication (physical.rs): PhysicalTable now stores both table: Table and extracted fields (table_name, network). While acceptable given Table appears immutable, this could be documented.

4. Logging change (parquet_writer.rs): The trace span changed from compact to full table reference - verify this is intentional.

Minor suggestions

• Consistency in using dataset_reference vs dataset.reference() in some files
• Removal of network() and schema() accessors from ResolvedTable changes the caller interface

Overall, this is a solid refactoring. The inline comments have specific suggestions for the items above.