Add indexed_dataset_version_time OTel gauge

[leonhandreke]

Tracks Unix timestamp of max last_seen across all documents per dataset, set after each successful catalog refresh. We'd ideally use last_export for this, but that's only available in the catalog index before the new index is loaded — once the catalog refresh completes, it's gone. So we derive freshness by max-aggregating last_seen across all indexed documents instead.

Changes

• SearchProvider.get_index_max_date() — new method on the base class that runs a max aggregation on any date field; works for both ES and OpenSearch via the existing search() abstraction
• yente/data/metrics.py — new module holding the gauge and update_dataset_version_metric()
• indexer.py — calls the update function after alias rollover completes
• tests/test_search_provider.py — covers empty index (→ None) and the max-of-two case

🤖 Generated with Claude Code

#1079

[leonhandreke]

@pudo how much do you hate this mechanism and do you have a better idea? Basically, I want to have a metric that makes it possible to alert on "default older than X". Right now, we can only export stale/not stale (index_version == catalog.version). That's another option, to just alert on "stale longer than X hours", but not as elegant. WDYT?