fix(loader): handle BaseMetadataConfig in serialize_value

[MicaPerdomo]

Summary

Fixes #38 — filters not assigned to retrieval tools during migrations.
Fixes #39 — collect_classes keying retrieval tools by __name__ instead of name attr.

Problem

When a user defines filters on a retrieval (filters = [GenreMetadata, LanguageMetadata]), those filters never reach the platform. The issue had three layers:

1. Serialization: serialize_value didn't recognize BaseMetadataConfig subclasses, falling back to repr() and producing "<class 'data.movies.metadata.GenreMetadata'>" instead of "genre".
2. Ordering: _sync_content_with_api created retrievals before metadata configs, so filter IDs didn't exist at resolve time.
3. Resolution: _set_if_defined("filters") copied raw string values to the API payload without resolving them to remote metadata config IDs.

Changes

• cogsol/core/loader.py: Add BaseMetadataConfig to type handling. Consolidate 6 identical issubclass branches into a single _SERIALIZABLE_TYPE_BASES tuple check. Fix collect_classes to key retrieval tools by name attr instead of __name__.
• cogsol/management/commands/migrate.py: Move metadata config upsert before retrievals. Replace _set_if_defined("filters") with proper ID resolution from new_remote["metadata_configs"], with type validation and descriptive error messages.
• tests/test_loader.py: Add tests for BaseMetadataConfig serialization and collect_classes keying behavior.

Test plan

• pytest — all 98 tests pass
• Verified metadata config upsert order is correct (before retrievals)
• Verified filter resolution follows same pattern as formatter resolution

[MicaPerdomo]

Related to #38 — this PR fixes the serialization step (filters stored correctly in migrations), but the full fix also requires changes in migrate.py to send the resolved filter IDs to the API.

[MicaPerdomo]

Also fixes #39.

[MicaPerdomo]

Updated this PR to fully fix #38 (previously it only fixed the serialization step).

The problem: when a user defines filters on a retrieval (filters = [GenreMetadata, LanguageMetadata]), those filters never reach the platform. Three things were broken:

1. Serialization (loader.py)
serialize_value didn't have a branch for BaseMetadataConfig. When makemigrations ran, it fell back to repr() and produced ["<class 'data.movies.metadata.GenreMetadata'>"] instead of ["genre"].

2. Execution order (migrate.py)
During migrate, the code created retrievals first and metadata configs after. So even with correct filter names, the metadata config IDs wouldn't exist yet when the retrieval tried to reference them.

3. ID resolution (migrate.py)
_set_if_defined("filters") just copied the raw string values into the API payload. It never resolved filter names to their remote metadata config IDs — unlike formatters, which already had proper ID resolution.

This update fixes all three layers: serialization recognizes BaseMetadataConfig, metadata configs are created before retrievals, and filter names are resolved to remote IDs (following the same pattern as formatters).