QA: Run and attach GPU/backend report for closed AI sprint features

[ChuxiJ]

Problem

Closed AI sprint features do not yet have a backend/GPU QA report attached. The story matrix still marks prompt-to-audio generation as blocked by environment, and Playwright tests mainly mock/store-check generation instead of validating real ACE-Step API behavior.

Root Cause

AI feature implementation and lightweight UI tests landed before the GPU integration environment was revalidated.

Solution

• Run the GPU/backend integration suite for prompt generation, negative prompt, prompt library apply, queue retry, voice-conditioned generation, cover/repaint, and audio/style influence.
• Capture logs, request/response metadata, generated asset evidence, and known failures.
• Update the story matrix from blocked/unknown to pass/fail/deferred.

Verification

• Run the documented GPU integration commands in tests/gpu-integration.
• Attach report artifacts and screenshots/audio metadata to this issue.
• File follow-up bugs for any failing backend workflows.

Files to Touch

• tests/gpu-integration/*
• docs/qa/story-matrix.md
• tests/e2e/generation-panel.spec.ts
• src/services/generationPipeline.ts

[m13v]

the failure mode here is store-check tests passing forever while the real generation pipeline silently breaks. mocked Playwright assertions confirm UI state transitions but never see a 502 from the inference backend, never catch schema drift in the prompt-to-audio response, never notice p99 latency walked off a cliff after a model swap. the only way out is splitting into two suites: a fast mocked path that owns UI regression, and a real-gpu path that owns ground truth. mixing them is what produced the blocked/unknown rows on the matrix in the first place.