Upgrade to EdgeQuake v0.26.5

From: v0.26.4 · To: v0.26.5 · CD: GHCR (edgequake, edgequake-frontend, edgequake-postgres)

Ops/product patch: SPEC-145 Langfuse generation / LLM observation I/O is complete by default (full prompt + completion; no product-side length truncation). Secrets remain redacted; ingest document content stays Preview. No new migrations — schema train remains 149 from upgrade-to-0.26.0.md.

Highlights

Area	What changed
Langfuse I/O	Generation Input/Output = full LLM payload (Complete class)
Default ceiling	`EDGEQUAKE_LANGFUSE_IO_MAX_BYTES=0` (unlimited); optional positive clamp
Stream	Generation span held until tokens end; I/O recorded once assembled
Helm	`api.langfuse.ioMaxBytes` → ConfigMap

Restart the API after deploy so new traces pick up Complete I/O (frontend image unchanged for this cut).

Sequence

1. Pull GHCR images for 0.26.5 (especially edgequake API)
2. Deploy v0.26.5 API + frontend (no migrate step — schema still 149)
3. Verify /health and OpenAPI versions are 0.26.5
4. Run a long Mix/query and confirm Langfuse generation Input/Output are complete

Distroless API note

Do not docker exec … curl inside the API container — there is no shell or curl. Probe from outside:

curl -s http://localhost:8080/health

Compose / quickstart pin:

EDGEQUAKE_VERSION=0.26.5 docker compose -f docker-compose.quickstart.yml up -d

Kubernetes:

EDGEQUAKE_VERSION=0.26.5 make k8s-install
# or set global.edgequakeVersion: "0.26.5" in values

Verify

curl -s http://localhost:8080/health | jq -r '.version'   # expect 0.26.5
curl -s http://localhost:8080/api-docs/openapi.json | jq -r '.info.version'  # 0.26.5
# Optional: make spec145-langfuse-e2e (live OTLP + tail marker)

Out of scope in this cut

New schema / migrate step (train stays 149)
Fresh Acc n=200 medical-mid run (attested existing publish/latest)
Dumping ingest document bodies into Langfuse

Detail: specs/145-fix-truncated-logs/ · Observability: docs/OBSERVABILITY.md.