What this PR does / why we need it:

The monitoring UI Freshness column previously used metric_date (the DQM computation window start) as a proxy for data freshness. This was misleading — metric_date reflects when metrics were computed, not how fresh the underlying data is.

This PR introduces a new max_event_timestamp field that stores MAX(event_timestamp) from the actual data source. The UI now displays age relative to the newest source event, giving an accurate picture of data staleness.

Changes

Backend — monitoring_service.py

• Added _as_utc_datetime() helper to normalize timestamp-like values to timezone-aware UTC datetimes.
• Added _newest_event_in_window() to clamp the source's max timestamp to the computation window.
• All metric save paths (_save_computed_metrics, _save_log_metrics) now propagate max_event_timestamp.
• Feature service metrics aggregate max_event_timestamp across contributing feature views.

Backend — monitoring_utils.py

• Added MONITORING_TIMESTAMP_FIELDS constant to DRY-up timestamp field lists across offline stores.
• Added max_event_timestamp to all three column definition lists (feature, feature_view, feature_service).
• normalize_monitoring_row now handles pandas NaT/NaN for the new field.

Schema — all offline stores (BigQuery, Snowflake, Redshift, Postgres, Spark, Oracle, Dask, DuckDB)

• Added max_event_timestamp column to CREATE TABLE DDLs for all three monitoring tables.
• Added ALTER TABLE … ADD COLUMN migration for existing tables.

UI — FeatureMetricsTable.tsx / useMonitoringApi.ts

• Freshness column now uses max_event_timestamp with a fallback to metric_date.
• Fixed render function signature ((_val, item)) so the full FeatureMetric object is passed to the freshness logic.
• Added Number.isNaN guards for robustness.
• Updated tooltip to show "Newest event" instead of "Data from".

Docs

• Updated feature-monitoring.md with max_event_timestamp in the example response and clarified the freshness semantics.

Which issue(s) this PR fixes:

Fixes inaccurate freshness display in the monitoring UI — the column previously showed metric_date (computation window) instead of actual data recency.

Checks

• I've made sure the tests are passing.
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests
• Integration tests
• Manual tests

Tests added

• tests/unit/monitoring/test_feature_freshness.py — unit tests for _newest_event_in_window (inside window, clamped, before window) and end-to-end auto_compute verification that max_event_timestamp is persisted.
• tests/integration/monitoring/test_monitoring_integration.py — extended test_auto_compute_uses_pushdown_for_max_timestamp to verify max_event_timestamp in saved metrics.
• Manual — verified on a live cluster with Dask offline store that freshness correctly reflects MAX(event_timestamp) from source parquet files.

Misc

The new column is nullable and added via ADD COLUMN IF NOT EXISTS migrations, so this is backward-compatible with existing monitoring tables.