feat: File-based checkpoint persistence for connectors (default) #107

Problem

Connector poll state (cursors, etag, last-seen timestamps, PR cache) is held in memory. On service restart, connectors lose their position and may:

• Re-process events in the lookback window (causing duplicate route firings)
• Miss events created between the last poll and the restart
• Reset the PR cache, causing re-evaluation of unchanged PRs

Proposal

Make file-based checkpoint persistence the default for all connectors.

Requirements

• Connectors persist their poll checkpoint to a file after each successful poll
• On startup, connectors read the checkpoint file and resume from where they left off
• Checkpoint files live in a configurable directory (default: .orgloop/checkpoints/ relative to the module directory)
• Each source gets its own checkpoint file: <source-id>.json
• If no checkpoint file exists (first run), connector uses initial_lookback config (existing behavior)
• File writes are atomic (write to temp, rename) to prevent corruption on crash
• This should be the default — no config required to enable

Configuration (optional overrides)

defaults:
  checkpoint:
    store: file          # default (was: memory)
    dir: .orgloop/checkpoints

Context

After restarting orgloop to switch repos, the GitHub connector's in-memory checkpoint was lost. Combined with etag caching behavior, this caused new events to not be picked up until the connector's lookback window expired.