Fix Speaches STT and custom TTS voices

[drascom]

Please describe the changes in your PR. If it is addressing an issue, please reference that as well.

What changed

This updates the Speaches adapters in pipecat so they match a stock Speaches server without requiring server-side compatibility shims.

Why

Speaches exposes OpenAI-compatible HTTP endpoints for both transcription and speech generation, but the previous adapters had two integration mismatches:

• STT expected a custom websocket path instead of using /v1/audio/transcriptions
• TTS mapped voice through OpenAI's fixed voice whitelist, which breaks custom Speaches voices such as fettah

Impact

Dograh deployments pointing at a Speaches server can now:

• use Speaches STT via the existing OpenAI-compatible transcription endpoint
• use custom Speaches TTS voice IDs directly

Root cause

The Speaches adapters were inheriting behavior tailored to Dograh/OpenAI transport assumptions rather than Speaches' actual OpenAI-compatible API surface.

Validation

• python3 -m py_compile on the updated adapter and test files
• Added focused regression tests for Speaches STT request shaping and custom TTS voice passthrough

Companion Dograh PR: <DOGRAH_PR_URL>

Related discussion: <DISCUSSION_URL>

This PR contains the adapter-level Speaches fixes. The Dograh PR depends on this landing first because it updates the service wiring and submodule pointer.