Restore ConversationalAIAPI compatibility and render-mode fallback

[alienzh]

Summary

• Restore the ConversationalAIAPI export and compatible enum/type aliases used by the Shengwang Web implementation.
• Align IConversationalAIAPIConfig with the top-level rtmEngine API.
• Add configurable WORD-to-TEXT transcript fallback when word timing data is unavailable.
• Preserve rendered transcript history while discarding pending WORD queue data.
• Bump the core and React packages to 2.9.1.

Migration

Replace the previous nested RTM configuration:

rtmConfig: { rtmEngine }

with:

rtmEngine

enableRenderModeFallback defaults to true. Set it to false to retain WORD-only behavior when timing data is missing.

Validation

• pnpm test — 120 core and 33 React tests passed
• Core typecheck passed
• React typecheck passed
• Interop typecheck passed
• Lint passed
• Production build passed