Pipeline observability: align web/digest with scheduler run state

Problem

Observability surfaces are still wired to legacy pipeline_runs data, but the scheduler now tracks active runs in scheduler-state.json and agentd logs.

Current mismatch:

• /pipeline web API/dashboard reads pipeline_runs from state.db
• pipeline digest email also reads pipeline_runs
• scheduler run loop does not insert/update pipeline_runs

Observed symptom:

• dashboard reports stale active run(s) from historical records
• digest metrics remain tied to old data instead of current scheduler activity

Impact

• Operators receive incorrect health signals
• "active runs" can be ghost entries
• daily digest can report misleading shipped/failure counts

Proposed fix

• Define a single source of truth for scheduler run state
• Either:

1) write scheduler lifecycle events into pipeline_runs again, or 2) migrate web/digest to scheduler-native state + agentd run history

• Add cleanup/migration for stale legacy running rows

Acceptance criteria

• Dashboard/digest reflect current scheduler activity
• No stale/ghost active runs after migration