[R3] Registered Prometheus metrics never incremented (dead series)

[DsThakurRawat]

Labels: bug, observability, priority:medium
File: src/tether/observability/prometheus.py — tether_denoise_steps_total/inc_denoise_steps (:90/:293), tether_episodes_active/set_episodes_active (:162/:457), tether_model_swaps_total/inc_model_swap (:143/:435)

Description. These metrics are defined and exported but have no call site (model-swap is only referenced in a comment).

Why it matters. Operators build Grafana panels/alerts on denoise throughput, active-episode count, or hot-swap events and get permanently-empty series.

Tasks.

• Wire the increments at their intended call sites (denoise loop, episode tracker, model-swap completion) or remove the metrics.

Dedup. #50/#99/#42/#258 wire other metrics (cache hit rate, drift score, circuit-breaker, cache bytes). These three are not among them — confirm cache_hit/miss aren't subsumed by #50 before including them.