feat(voice): improve assistant voice input with provider-based architecture

[dongaiqiang]

Background

This PR improves voice input stability and maintainability in Happy App.

What changed

• Consolidated voice entry behavior to avoid conflicting button states
• Refactored voice input logic into a provider-based structure
• Added strategy-based mode boundary for future extension
• Added tests for voice mode decision logic

Scope

• packages/happy-app/sources/components/AgentInput.tsx
• packages/happy-app/sources/features/custom-asr/SmartVoiceButton.tsx
• packages/happy-app/sources/features/custom-asr/useCustomASR.ts
• packages/happy-app/sources/features/voice-input/*

Validation

• yarn workspace happy-app test run sources/features/voice-input/providers/providerFactory.test.ts ✅ (9/9 passed)
• Manual test on mobile completed: streaming voice input updates continuously and send/voice behavior is correct
• yarn workspace happy-app typecheck currently reports additional errors on top of latest upstream/main (outside this PR’s core change path)

Notes

• No intentional changes to unrelated RPC flow
• External UX remains consistent while internals are decoupled for future extension

[dongaiqiang]

Thanks for reviewing.

This PR focuses on voice input reliability + provider-based decoupling:

• single entry behavior
• provider/strategy boundary
• tests for mode decision logic

I kept unrelated RPC flow untouched.
Happy to split into smaller PRs if preferred.