feat(transform-dedup): Add file-based persistence for dedup state

[warren-t-c]

Problem

@orgloop/transform-dedup (v0.1.10) uses an in-memory Map for dedup state. On service restart, the seen map is cleared, causing previously-deduplicated events to pass through again and trigger duplicate route firings.

Current behavior

From the README:

"State is in-memory only and lost on restart."

Config schema: store: string — Storage backend. Only "memory" currently.

Any service restart (config change, update, crash recovery) loses all dedup state, causing events within the dedup window to be re-delivered.

Proposal

Add store: "file" option, using the same FileCheckpointStore infrastructure from #107.

Suggested config

transforms:

• name: dedup
  type: package
  package: "@orgloop/transform-dedup"
  config:
  window: "15m"
  store: "file"
  fields:
  - "type"
  - "source"
  - "provenance.platform_event"
  - "provenance.issue_number"
  Requirements

1. Persist hash + timestamp to file on each pass/drop decision
2. On startup, load persisted hashes, filter out entries older than window
3. Atomic file writes (write to temp, rename) — same as connector checkpoints
4. Files in .orgloop/checkpoints/dedup-.json
5. store: "memory" remains default for backward compat
6. Periodic expired-entry cleanup (existing timer pattern)

Context

Related: #107 — extends connector checkpoint persistence to the dedup transform.