feat(query): migrate to ClickHouse + web-Lapache backend endpoints

[jacobjurek]

Summary

Migrates the query service from Postgres to ClickHouse, and brings the query work it builds on up to date on main. Scoped to query/ plus the query service's docker-compose block and the CLICKHOUSE_* entries in example.env — no other service is touched.

Note: the ClickHouse migration sits on top of the web-Lapache backend endpoints (clusters, date selector, decimation, signal-names), which were on the feature branch but not yet on main. Those files (e.g. service/cluster.py) don't exist on main, so the migration can't be merged in isolation — this PR includes them as one coherent unit.

ClickHouse migration

• Drop sqlalchemy + psycopg2; add clickhouse-connect (HTTP interface, port 8123).
• connection.py: one shared client; init_db creates the service's own metadata tables in ClickHouse — query_log (MergeTree, append-only), query_token + signal_definition (ReplacingMergeTree, read with FINAL).
• Signal/cluster SQL ported to ClickHouse: DISTINCT ON → LIMIT BY, EXTRACT(EPOCH …) → integer bucketing on the timestamp micros column, AT TIME ZONE …::date → toDate(produced_at, tz). produced_at is a MATERIALIZED column (queryable, just hidden from SELECT *), so it's always named explicitly.
• Models → plain dataclasses; metadata services use clickhouse-connect.
• docker-compose query block + example.env: CLICKHOUSE_* over HTTP 8123 (gr26 keeps native 9000).

Supporting endpoints (pre-existing on the feature branch)

/signals/names, /clusters, /clusters/dates; raw-data decimation via max_points; cached highest-frequency anchor signal; dev reload watcher scoped to the app package.

Verification

• pytest → 12/12 pass.
• Exercised end-to-end against ClickHouse: /signals (pivot + decimation via LIMIT BY), /signals/names, /clusters (gap split), /clusters/dates (tz), /token (write), token-auth on /signals (FINAL read), and query_log written on every /signals call.

🤖 Generated with Claude Code