stats pipeline: aggregator, stats storage, and dashboard tab

[aaylward]

turn shipped logs into served stats. part of #1365.

• aggregator: interval-loop worker (one_d4_worker pattern) computing per-entity aggregates — iili url popularity, per-tool mcp usage, bot traffic share, geo — into a stats db on shared_postgres (new one-shot *_db_init service; mind the 0.25cpu/256m cap on that instance)
• small stats service in front of the tables
• new tab in the metrics-systems dashboard next to the prom_proxy views, same registry-driven shape

division of labor: prometheus keeps low-cardinality 7-day operational series; this owns high-cardinality, long-retention, per-entity aggregates that don't belong in a tsdb.

[aaylward]

carrying over from #1459: 'popular chessql queries' can't come from access logs (query text rides in POST bodies, and mcpserver's tool-call log deliberately excludes arguments). needs a deliberate app-level event with the normalized query text — decide whether/where to emit it when building the aggregator.

[aaylward]

resolved: no new event needed. one_d4's QueryController already logs the full query at INFO (query={} limit={} offset={} player={}), and with the json encoder the query text is machine-readable as arguments[0] of that line — LogbackConfigTest pins the template-plus-arguments shape. so 'popular chessql queries' becomes a query over shipped app logs once source=app shipping lands; the aggregator in #1463 covers the caddy-derived stats.