feat(research): persist ranking observations

[hubab1]

Summary

• add an app-only, generation-safe ranking workflow for pre-live keyword research projects
• persist complete 200-result shared ranking crawls without creating fake tracked apps, tracked keywords, ranks, or refresh statuses
• make shared and tracked ranking persistence independently monotonic, deterministic, and rollback-safe
• rebuild shared keyword statistics in the owning transaction and schedule metadata enrichment only after a successful commit
• preserve current App Store Web versus iTunes catalog, language, metadata, rating, and canonical-storefront provenance
• protect shared UI model contexts from committing or rolling back unrelated pending edits
• route coordinator-backed MCP persistence through the same full-result and post-commit rules

Safety and scope

• no schema change
• no credential or session handling
• no UI surface
• stale project/keyword incarnations, cancellation, provider failure, partial pages, delayed sources, equal retries, and batch rollback are covered
• metadata freshness remains independently monotonic for global app facts, canonical storefront presentation, storefront detail records, screenshots, and ratings

This progresses #3 but does not close it.

Validation

• 16/16 dedicated research-ranking workflow tests passed
• 101/101 ranking, MCP, and metadata compatibility tests passed
• full OpenASO suite passed: 510 tests (514 parameterized test invocations)
• project-file lint and whitespace checks passed