Time-series metrics (Prometheus-style) for live factory data

[ApocDev]

Far-future, not scheduled.

Today the bridge pushes live state as replace-snapshots (state.stats every ~10s, state.built, research, TURD). A Prometheus-style evolution would retain history instead of overwriting, so trends can be graphed and alerted on.

Sketch:

• Treat the ~10s stats push as a scrape interval and persist samples (good, produced/s, consumed/s, tick, timestamp) into a time-series store rather than the single production_stats row-per-good snapshot. Down-sample and retain windows so it does not grow unbounded.
• Same for built-machine counts and power, anything currently snapshot-only.
• Web: time-series charts (throughput over time, per item and machine) and threshold alerts ("iron plate output < X/s", "block starved for N minutes").
• Alternatives to weigh: a real Prometheus exporter endpoint the user scrapes with their own stack, versus an embedded lightweight TSDB (for example SQLite rollups) owned end-to-end. The mod side barely changes, since it already emits the right counters.

This builds on the live production stats and the state.* bridge messages.