feat(backend): implement async job flow and real gRPC integration

[ElifFCebeci]

July 21-22-23 Updates: API Gateway Async Job Flow, Real gRPC .

---

Description

Summary

• /enhance and /enhance/batch now run a real async job flow instead of a fake 202 response: QUEUED → RUNNING → COMPLETED/FAILED in Redis, result is queried via GET /enhance/{job_id}.
• /health and /models are now wired to real Health / ListModels gRPC calls (static mock removed).
• gRPC client: 4-connection round-robin pool + retry/backoff only for UNAVAILABLE.
• CORS, Prometheus /metrics, route-based upload size limit (413), graceful shutdown (Redis/gRPC won't close on SIGINT/SIGTERM until background jobs finish).
• api/openapi.yaml generated; JobStatus comment in common.proto updated to match the async model.
• Go 1.25.12 + quic-go v0.59.1 — closes 17 vulnerabilities reported by govulncheck.

---

Out of Scope

• services/ml/inference_server (Python gRPC server) is not included in this PR and has not been written yet — since the gateway currently has no counterpart, /health, /models, and jobs will return "unreachable/failed". This is expected.

---

Test Plan

• go build ./..., gofmt -l ., go vet ./... — clean
• Manual smoke test with Redis + real server: /enhance → 202 → job recorded, 404 for non-existent job, 413 for files over 25MB, correct error message when gRPC is absent for /health + /models, CORS with allowed/denied origin
• Graceful shutdown verified via code review; real SIGTERM signal could not be live-tested on Windows/local sandbox (expected to work correctly on Linux/k8s)