Voice mode: allow tap-to-toggle without auto-submit (autoSubmit: false for tap mode)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Tap mode's ergonomics are great — tap to start, speak hands-free, tap to stop — and for many people they're strictly better than holding a key for the entire utterance. But today the second tap always sends (for transcripts ≥3 words), with no way to disable it:

• mode: "tap" is documented and implemented as "tap to start, tap to stop+submit."
• autoSubmit is explicitly hold-mode only ("Submit the prompt when hold-to-talk is released (hold mode only)").

So the two built-in modes force an unwanted trade-off:

┌──────┬───────────────────────────┬────────────────────────────┐
│ │ Hands-free while speaking │ Review/edit before sending │
├──────┼───────────────────────────┼────────────────────────────┤
│ hold │ ❌ (must hold the key) │ ✅ │
├──────┼───────────────────────────┼────────────────────────────┤
│ tap │ ✅ │ ❌ (always auto-sends) │
└──────┴───────────────────────────┴────────────────────────────┘

There's no combination that gives both hands-free dictation and a chance to review — which matters a lot for dictation, since transcripts frequently need a quick fix (a misheard word, a stray "um") before they're worth sending. Auto-send on every tap means either living with mistakes or fighting the tool.

Proposed Solution

What I'm requesting

Let tap mode insert the transcript and wait for the user to press Enter, instead of always submitting on the second tap. Concretely: make the existing voice.autoSubmit setting apply to tap mode, so {"voice": {"mode": "tap", "autoSubmit": false}} records hands-free but leaves the text in the prompt for review.

Proposed behavior

• Honor voice.autoSubmit in tap mode.
• autoSubmit: true (or unset, to preserve today's default) → current behavior: tap-to-stop submits.
• autoSubmit: false → tap-to-stop inserts the transcript at the cursor and waits for Enter, exactly like hold mode's default.
• Update the autoSubmit description from "(hold mode only)" to cover both modes.

Alternative Solutions

Alternatives considered

• Hold mode — solves the review problem but defeats the point of hands-free dictation (you hold the key the whole time you talk).
• External STT tools (Superwhisper, etc.) configured not to press Return — works, but requires leaving the built-in voice feature entirely for what should be a one-line setting.

Priority

High - Significant impact on productivity

Feature Category

CLI commands and flags

Use Case Example

I would now be able to use tap mode for dictation all the time without worry and with editing. I wouldnt have to switch to hold, which sucks.

Additional Context

Additional context

Tap mode itself landed from community demand (#33787). This is the natural follow-on: the plumbing (autoSubmit) already exists — it just doesn't apply to tap mode yet.

Verified against Claude Code v2.1.197.