Hybrid voice mode: tap-when-empty, hold-when-typing

Feature Request

Current behavior:

• tap mode: Space toggles recording (only works with empty input)
• hold mode: Hold Space to record

Desired behavior:
Combine both - when input empty, tap Space to toggle. When input has text, hold Space to dictate inline.

Use case:
Often start with voice (tap), then want to add more via dictation after typing corrections. Currently must switch modes manually with /voice hold.

Proposed UX:

• Empty input → tap behavior
• Has text → hold behavior
• Single mode name: auto or hybrid