* Add ClickHouse schema metrics (schema_metrics + view refresh)

Adds per-table size gauges via ClickhouseTableMetrics job class and
per-view refresh status/timing gauges collected inline in check().
Introduces schema_metrics config block and _VIEW_REFRESHES_QUERY
with graceful handling for ClickHouse < 24.3 and missing permissions.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(clickhouse): rename changelog entry to match PR number 23900

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(clickhouse): sync SchemaMetrics model formatting with spec.yaml

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(clickhouse): gate view refresh metrics on schema_metrics.enabled

_collect_view_refresh_metrics was running on every check cycle when dbm
was enabled, regardless of schema_metrics.enabled (which defaults to false).
Gate it on self.table_metrics so it respects the opt-in config.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* fix(clickhouse): avoid duplicate db: tag on schema metrics

Per-table and per-view schema metrics appended `db:<database>` on top of the
instance-level `db:<connection_db>` base tag, producing two conflicting `db:`
tags for any entity outside the connection database. Strip the base `db:` tag
before adding the entity's own database, mirroring the postgres
`tags_without_db` pattern, so each series carries exactly one `db:` tag.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* Address review feedback: null guard, feature flag, constant naming, tests

- Remove redundant schema_metrics null check (always initialized with defaults)
- Register FeatureKey.SCHEMA_METRICS in config.py feature reporting
- Rename _DEFAULT_COLLECTION_INTERVAL → DEFAULT_COLLECTION_INTERVAL per AGENTS.md
- Add restart hint to permission-denied log message
- Add unit tests for _collect_view_refresh_metrics() and _handle_view_refreshes_error()

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Move view refresh logic into ClickhouseTableMetrics

_collect_view_refresh_metrics and _handle_view_refreshes_error now live
on ClickhouseTableMetrics and run inside run_job(), sharing the same
collection interval and DBMAsyncJob lifecycle as table size gauges.
State flags and query/status-map constants move to table_metrics.py.
clickhouse.py loses the duplicate if-table_metrics block, the two
module-level constants, the class attribute, and the three init flags.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Fix run_job() table-size tests silently exercising view refresh path

_patch_query now patches execute_query_raw to return [] alongside
_execute_query, so the view refresh half of run_job() succeeds cleanly
instead of swallowing a connection error. The cluster routing test also
patches execute_query_raw and gains an assertion that view refresh
queries go through clusterAllReplicas too.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Fix ruff formatting: collapse single-line format() call in table_metrics.py

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Restore trailing spaces in conf.yaml.example stripped by ruff

ddev validate config requires spec-generated trailing spaces on
'default: ' and 'For example: ' comment lines.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Fix view_refresh metrics to emit per-replica service checks

In single_endpoint_mode, clusterAllReplicas returns one row per replica per
view. The previous seen dedup keyed on (database, view) silently discarded
all but one replica's status, which could hide per-replica Error states.

Now keys seen on (database, view, host) and tags each series with host: so
each replica surfaces its own refresh status independently.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Use LIMIT 1 BY in table sizes query instead of Python-side dedup

clusterAllReplicas returns one row per replica per table; pushing the
dedup into SQL with LIMIT 1 BY database, name is cleaner than filtering
in Python.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Fix test_table_metrics for host tag and SQL-side dedup changes

- Add host field to _view_refresh_row fixture (query now returns hostName())
- Assert host: tag appears in view refresh service check
- Replace Python-side dedup test with assertion that LIMIT 1 BY appears in
  the table sizes query, since dedup is now the SQL's responsibility

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Use LIMIT 1 BY in view refreshes query instead of Python-side dedup

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Remove stale Python-side dedup test for view refreshes

Deduplication is now enforced by LIMIT 1 BY in the SQL query.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Fix ruff: remove extra blank line in test_table_metrics.py

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Remove view.refresh service check in favour of view.refresh.status gauge

Service checks are slated for deprecation; new ones should not be added.
The view.refresh.status gauge already covers the same OK/WARNING/CRITICAL/UNKNOWN
signal, so the service_check() call is redundant.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>