Simplify dictate to auto-start recording and exit after one utterance

[alexkroman]

Summary

Refactor assembly dictate to always auto-start recording and exit after one utterance, removing the interactive idle-loop mode and the --once flag's functional behavior. This simplifies the UX for the common case (pipe-friendly single utterance) and makes it the default.

Key Changes

• Auto-start recording: Recording now begins immediately when dictate runs, eliminating the "Press Enter to start" prompt and idle-loop mode
• Single-utterance default: The session captures one utterance and exits, closing stdout pipes so downstream commands unblock (e.g., assembly dictate | assembly llm …)
• Deprecate --once: The flag is now a hidden no-op that warns users it can be omitted; kept only for backward compatibility with existing scripts
• Simplify key handling: Replaced TOGGLE_KEYS / QUIT_KEYS distinction with unified STOP_KEYS (Enter, Space, q, Esc, Ctrl-D all stop recording); removed the idle-prompt quit-key logic
• Remove _session() function: Eliminated the branching logic that chose between single-shot and interactive modes; _capture_and_transcribe() is now called directly
• Remove stdio.stdout_is_tty() check: No longer needed since single-utterance is always the behavior
• Update help text and examples: Clarified that recording starts immediately and dictate exits after one utterance; removed --once from visible options and examples

Implementation Details

• The seams fixture no longer patches stdio.stdout_is_tty() since TTY detection is unused
• Test key sequences simplified: removed leading toggle keys and trailing quit keys (e.g., ["\r", None, "\r", "q"] → [None, "\r"])
• Keyboard timeouts now only include zero-timeout in-recording polls (no blocking None waits for idle prompts)
• Quit keys during recording now transcribe the captured audio instead of discarding it
• The --once deprecation warning is suppressed by --quiet to avoid noise in automated scripts

https://claude.ai/code/session_01RFDkryY7tiJMLs93ednr1G