Closes #11

Summary

Adds pkg/batch — an in-memory async job processor for large deduplication workloads.

API

# Submit
POST /v1/batch
{"chunks":[...],"options":{"dedup":{"enabled":true},"compress":{"enabled":true}}}
→ {"job_id":"batch_1234","status":"queued"}

# Poll
GET /v1/batch/batch_1234
→ {"status":"processing","progress":0.45,"started_at":"..."}

# Results
GET /v1/batch/batch_1234/results
→ {"chunks":[...],"stats":{"original_tokens":5000,"final_tokens":1200,...}}

pkg/batch

• Processor: configurable worker pool (default 4), job queue (default 1000), result TTL (default 24h)
• Submit: enqueues job, returns immediately
• Get: current job state (queued / processing / completed / failed)
• Results: deduplicated chunks + pipeline stats for completed jobs
• List(status): filter jobs by status
• evictLoop: background goroutine removes expired results every 5 minutes
• Stop: graceful shutdown

HTTP handlers are in cmd/api_pipeline.go (part of feat/4-pipeline).

Files

• pkg/batch/batch.go
• pkg/batch/batch_test.go — 7 tests