fix: Voice-conditioned generation should fail clearly when selected voice audio is missing

[ChuxiJ]

Problem

Voice-conditioned generation can silently proceed without voice audio. If loadAudioBlobByKey returns no blob, generation continues without a user-visible warning. addVoice can also return before IndexedDB persistence finishes, creating a race window.

Root Cause

Voice profile metadata and voice audio blob persistence are not treated as an atomic precondition for voice-conditioned generation.

Solution

• Await/confirm audio blob persistence before a voice profile is considered ready.
• Fail voice-conditioned generation with an actionable error if selected voice audio is missing.
• Add recovery UX to re-upload/re-record the missing voice sample.
• Add tests for missing blob and persistence race cases.

Verification

• Add store/service tests for addVoice persistence completion.
• Add generation tests for missing selected voice audio.
• Add UI test for the user-facing error and recovery path.

Files to Touch

• src/store/voiceStore.ts
• src/services/generationPipeline.ts
• src/services/voiceUploadService.ts
• src/components/voice/*