fix: harden orchestration state recovery

[chihsuan]

Context

Rework dispatch could lose durable context or get stuck after transition failures, and multi-repo startup cleanup missed secondary repos.

TL;DR

Make rework state changes durable before dispatch and recover cleanly after restarts.

Summary

• Persist PR review rework context before moving Linear issues back to active.
• Let CI retries escalate after max attempts, while keeping active-run protection.
• Clear failed CI dispatch markers so the same SHA can be retried.
• Interrupt persisted running runs for every configured repo on startup.
• Reject repo keys that collide after workspace path sanitization.

Alternatives

• Keep retrying later without ordering changes; rejected because dispatch can run first with missing context.
• Keep workspace cleanup scoped by raw repo key only; rejected because sanitized paths still collide.

Test Plan

• make all
• mix test test/symphony_elixir/ci_poller_test.exs test/symphony_elixir/pr_review_poller_test.exs test/symphony_elixir/config_split_test.exs test/symphony_elixir/orchestrator_status_test.exs
• mix specs.check