feat: queue-aware speak() with enriched return metadata (#4)

* feat: queue-aware speak() with enriched return metadata

Replace the barge-in-on-self behavior with a serial speech queue.
When speak() is called while audio is already playing, the new request
is queued and plays automatically when the current item finishes.
The return value now includes full queue state (position, currently
playing info, estimated wait, suggested actions) so the agent never
needs a separate status call to understand the output channel.

- Add SpeechQueue class with thread-safe enqueue/cancel/drain
- speak() returns "queued" status with enriched metadata when busy
- speech_status() always includes queue depth and currently_playing
- stop(job_id) can cancel a specific queued item or interrupt current
- stop() with no args interrupts current + clears entire queue
- Add 18 tests covering queue mechanics, return formats, edge cases

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: resolve ruff lint and format errors for CI

- Remove unused import `resolve_model` in http_api.py (F401)
- Remove unused variable `player` in server.py (F841)
- Remove unused import `pytest` in test_speech_queue.py (F401)
- Remove unused variable `original_drain` in test_speech_queue.py (F841)
- Fix unsorted import block in test_speech_status_no_jobs (I001)
- Apply ruff format to server.py, tests/test_bus_wiring.py, tests/test_speech_queue.py

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: add threshold attribute to Gate ABC for pyright compliance

The abstract Gate class was missing a threshold type annotation that
VoiceGate defines in __init__. Pyright flagged server.py accesses to
voice_module.gate.threshold as unknown attribute errors.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: chazmaniandinkle

http_api.py

@@ -24,7 +24,7 @@
 from pydantic import BaseModel, Field
 
 from bus import ModalityBus
-from engine import MODELS, generate_audio, get_loaded_engines, resolve_model
+from engine import MODELS, generate_audio, get_loaded_engines
 from modality import EncodedOutput, ModalityType
 from modules.text import TextModule
 from modules.voice import VoiceModule

modality.py

@@ -115,6 +115,8 @@ class ModuleState:
 class Gate(ABC):
     """Input gate — decides if raw input contains signal worth decoding."""
 
+    threshold: float
+
     @abstractmethod
     def check(self, raw: bytes, **kwargs) -> GateResult:
         """Check if this input should pass through to the decoder."""