refactor(metadata-db): decouple PhysicalTable from revision

[shiyasmohd]

• Split PhysicalTable into two structs:
  • PhysicalTable (logical table entity with dataset metadata and active revision pointer)
  • PhysicalTableRevision (storage revision with path and writer job)
• Compute active status at query time via EXISTS instead of denormalized column
• Rename get_by_id to get_active_by_location_id with typed error handling
• Add tests for inactive revisions, metadata, and writer details"

[LNSD]

Do we need this second migration? The active status is already derivable at query time from physical_tables.active_revision_id: either via (pt.active_revision_id = ptr.id) AS active in JOINs or an EXISTS subquery.

For example, the list query could be:

SELECT
    ptr.id,
    ptr.path,
    EXISTS(SELECT 1 FROM physical_tables WHERE active_revision_id = ptr.id) AS active,
    ptr.writer,
    ptr.metadata
FROM physical_table_revisions ptr
ORDER BY ptr.id DESC
LIMIT $1

The denormalized column + trigger adds schema surface and sync complexity for a JOIN that's cheap (small table, indexed FK). I'd lean toward dropping the second migration entirely and just computing active in the queries.

[shiyasmohd]

@LNSD Updated the PR. Removed the migration.

[LNSD]

@LNSD Updated the PR. Removed the migration.

You also removed dependencies from the providers-* crates that I will need in a follow-up PR 😅

[LNSD]

Please, check my comments 🙂

[LNSD]

LGTM ✅