Background Worker Schema & Infrastructure

[b08x]

Sequence: 1 (foundation — no dependencies)

Persistence layer for the async ingest track (#56, #59, #67). Issues like the Async Ingestion Queue (#56) and Dead-Letter Queue (#59) need a place to track in-flight job state that isn't the ingest_review_entries Postgres table (that table is for human review of low-confidence/unrecognized files — see Store::PgIngestReviewRepository — not job scheduling).

Architectural decision: use the pure-Ruby Ohm library mapped to the existing Redis instance (docker-compose port 6380, already provisioned — see AGENTS.md). Avoids pulling in ActiveJob or a heavyweight Postgres-polling queue.

Required work

1. Boot.rb wiring: add a require_redis: flag to SFL::Boot.call, following the exact same pattern as the existing require_llm:/require_db: flags (see lib/sfl/boot.rb) — opt-in, no side effects when unset, so specs and non-ingest commands are unaffected. Add REDIS_URL to .env/.env.example. Implement connect_redis(env) → Ohm.redis = Redic.new(url).
2. Ohm model: scaffold SFL::Store::IngestJob < Ohm::Model with attributes document_id, file_path, state (pending/pass_one/embedding/pass_two/done/failed), error_trace; indices on state and document_id.
3. Add gem "ohm", gem "redic" to the Gemfile.

Out of scope here

Anything that consumes IngestJob records (the worker loop is #67; retry/DLQ semantics are #59).

Depends on: nothing. Blocks: #56, #59, #67, #68.

[b08x]

Architectural Decision: We will use the pure-Ruby Ohm library mapping to our existing Redis instance (port 6380). This avoids pulling in ActiveJob or writing heavyweight Postgres polling queues. We will define an IngestJob Ohm model to track states (pending, processing, failed, done).