Reconcile stuck ingest_state=error for fully-indexed sessions

You are being dispatched from flower Brief #30: Reconcile stuck ingest_state=error for fully-indexed sessions

Recall pointer:
- Use recall_brief with id 30 for the full folder if you need provenance.

Target:
- project: flower (/Users/mikeferrara/Documents/code/flower)
- branch: flower/ingest-state-reconcile
- worktree: not specified
- kind: fresh

Current brief spec:
## Problem
Sessions **#406, #737, #3252** (and likely others) have `ingest_state=error` but their segments
are fully chunked and **INDEXED in Meili** (`meili_id` set on ALL chunks: 87/87, 2/2, 82/82) and
are **recall-searchable** (verified: #3252 returns 6 of its segments). So the CONTENT is fine —
the STATE is wrong: a downstream timeout/retry marked the session `error` after indexing already
succeeded. Impact is cosmetic/monitoring (session reads "error"); it is NOT a recall/embed gap.
(This corrects ops cycle-86's "Mode-2 embed gap / 0 chunks / dark to recall".)

## Fix
1. **Reconcile command** — `flower:reconcile-ingest-state`: find sessions with `ingest_state=error`
   whose segments are all chunked + indexed (every chunk has `meili_id`), and set them to
   `indexed`. Idempotent; dry-run default + `--force`. Verify it flips #406/#737/#3252.
2. **Pipeline fix** — find where `ingest_state` is set to `error` and ensure a session that
   finishes indexing is marked `indexed`, not left `error` by a downstream timeout/retry
   (success should win over a late failure). Add a regression test.

## Acceptance
- Reconcile command flips truly-complete `error` sessions → `indexed` (dry-run + force); verified.
- Pipeline no longer leaves a fully-indexed session as `error`.
- Tests; suite green. `Brief: #<id>` trailer.

## Provenance
ops cycle 86 Mode-2 report; orchestrator verified the chunks ARE indexed + searchable → it is a
state-reconcile issue, not an embed gap. Quick fix — operator: "push to a quick codex agent."

Recent/key trace events:
[1] participant_joined flower-orchestrator: (no body)
[2] note_added flower-orchestrator: Sessions like #406/#737/#3252 read ingest_state=error but are fully chunked+indexed+recall-searchable. Add a reconcile command (error->indexed when all chunks have meili_id) + fix the pipeline so a finished session is not left error by a late timeout/retry.
[3] plan_proposed flower-orchestrator: ## Problem
Sessions **#406, #737, #3252** (and likely others) have `ingest_state=error` but their segments
are fully chunked and **INDEXED in Meili** (`meili_id` set on ALL chunks: 87/87, 2/2, 82/82) and
are **recall-searchable** (verified: #3252 returns 6 of its segments). So the CONTENT is fine —
the STATE is wrong: a downstream timeout/retry marked the session `error` after indexing already
succeeded. Impact is cosmetic/monitoring (session reads "error"); it is NOT a recall/embed gap.
(This corrects ops cycle-86's "Mode-2 embed gap / 0 chunks / dark to recall".)

## Fix
1. **Reconcile command** — `flower:reconcile-ingest-state`: find sessions with `ingest_state=error`
   whose segments are all chunked + indexed (every chunk has `meili_id`), and set them to
   `indexed`. Idempotent; dry-run default + `--force`. Verify it flips #406/#737/#3252.
2. **Pipeline fix** — find where `ingest_state` is set to `error` and ensure a session that
   finishes indexing is marked `indexed`, not left `error` by a downstream timeout/retry
   (success should win over a late failure). Add a regression test.

## Acceptance
- Reconcile command flips truly-complete `error` sessions → `indexed` (dry-run + force); verified.
- Pipeline no longer leaves a fully-indexed session as `error`.
- Tests; suite green. `Brief: #<id>` trailer.

## Provenance
ops cycle 86 Mode-2 report; orchestrator verified the chunks ARE indexed + searchable → it is a
state-reconcile issue, not an embed gap. Quick fix — operator: "push to a quick codex agent."
[4] status_change flower-orchestrator: (no body)

Recommended linked context:
{
    "todos": [],
    "scratchpads": []
}

Execution notes:
- Treat the brief as the source of truth.
- Keep work scoped to this dispatch request.
- Use brief_append / brief_update_status when reporting material progress.
- Add a git commit trailer `Brief: #30` to every commit for this brief so flower can exact-link commits back to the brief.