Expose session-based ANR / App Hang rate in sessions API

Sessions API exposes ANR / App Hang rate only as a user-based metric (anr_rate(), foreground_anr_rate()). There is no session-based equivalent, so teams defining mobile SLOs on a session denominator must mix session-based crash metrics with user-based ANR metrics.

Current behavior

• anr_rate() / foreground_anr_rate() use uniqIf over the user set filtered by the abnormal_mechanism tag (anr_foreground, anr_background) — src/sentry/snuba/metrics/fields/snql.py
• In Relay, abnormal_mechanism is attached only to s:sessions/user@none, not to the session counter c:sessions/session@none:
  • SessionSessionTags carries only status + common tags; SessionUserTags carries abnormal_mechanism (relay-server/src/metrics_extraction/sessions/types.rs)
  • In the abnormal branch of extract_session_metrics, the session counter is tagged session.status = "abnormal" with no mechanism (relay-server/src/metrics_extraction/sessions/mod.rs)
  • test_extract_session_metrics_abnormal asserts the session counter's tag keys are exactly ["release", "session.status"]
• abnormal_mechanism is not a filterable/groupable dimension on sessions in the sessions or metrics API today

Proposed change

Phase 1 — Relay (prerequisite): Add abnormal_mechanism to SessionSessionTags and populate it in the abnormal branch of extract_session_metrics. Low cardinality (anr_foreground, anr_background). No SDK change required — the SDK already sends the mechanism on the session payload; Relay is not propagating it to the counter metric.

Phase 2 — Sentry server (after Relay):

• Register new MRIs e:sessions/session.anr_rate@ratio and e:sessions/session.foreground_anr_rate@ratio
• Add sumIf-on-counter snql functions filtered by abnormal_mechanism (mirroring crashed_sessions())
• Expose session-based fields in the sessions API (e.g. anr_rate(session) / foreground_anr_rate(session); naming TBD)

Acceptance criteria

• Sessions API returns a session-based ANR rate, queryable with groupBy=project,release and interval, on the same pipeline as crash-free sessions
• Value matches a manual sessions{abnormal_mechanism in (anr_foreground, anr_background)} / total sessions computation

Notes

• No historical backfill: session-based values accumulate only from the Relay deploy date
• iOS App Hang rate reuses the same abnormal_mechanism path, so the change benefits both Android ANR and iOS App Hang

Start in relay-server/src/metrics_extraction/sessions/types.rs and relay-server/src/metrics_extraction/sessions/mod.rs, including test_extract_session_metrics_abnormal, to trace session counter tags. Then inspect src/sentry/snuba/metrics/fields/snql.py and the sessions API metric registration. Done means the API exposes session-based ANR and foreground App Hang rates whose grouped values match the stated manual session calculation.