Wire PR engine into the daemon (built but unwired)

[Pritom14]

Problem

The PR observation subsystem is fully implemented and unit tested, but it is not wired into the running daemon:

• daemon constructs the Lifecycle Manager with lifecycle.New(store, nil) and a noopMessenger, so PR nudges (and human /send) are silently dropped in production.
• Nothing produces PR observations: service/pr.Manager and scm/github.Provider.Observe exist but no poller calls them. The polling loop was parked as a follow-up (referenced as Tracker observer loop: poll + ApplyTrackerFacts into LCM #35 in the provider docstring).

Net effect: a session can open a PR, CI can fail, and the agent is never nudged because the whole observation path is dark.

Goal

Make the PR engine live end to end in the daemon, with graceful degradation when no GitHub token is configured.

Approach (branch self-discovery)

domain.SessionMetadata already stores Branch, so a poller can self-discover each session's PR by its branch via the GitHub API. This avoids needing a hook-based PR-URL capture path.

1. Real runtime messenger backing both human /send and lifecycle nudges.
2. FindPRForBranch discovery primitive on the GitHub provider.
3. A PR poller (the parked loop) that discovers, observes, persists, and reacts.
4. Daemon wiring: provider + PR service + poller goroutine with lifecycle-managed shutdown.

Acceptance

• Daemon wires a real messenger, provider, PR service, and poller goroutine.
• No GitHub token: poller is disabled with a warning, daemon still starts.
• gofmt, go vet, go build clean; race tests green for all touched packages.