Add Soniox STT and TTS provider support

[parthamukherjeesnow]

Wires up pipecat's SonioxSTTService/SonioxTTSService in the service factories, documents worked examples in .env.example, and clarifies in llm_configuration.md that STT/TTS providers (Soniox included) are configured via EVA_MODEL__STT/TTS, not EVA_MODEL_LIST.

Feature added — Soniox STT/TTS support, committed as f743bf4 "Add Soniox STT and TTS provider support":

• src/eva/assistant/pipeline/services.py
  • Imported SonioxSTTService/SonioxTTSService from pipecat.
  • New "soniox" branch in create_stt_service: forwards Settings fields present in params, converts language_hints from raw BCP-47 strings to pipecat Language enums (required by Soniox's dataclass).
  • New "soniox" branch in create_tts_service: same forwarding pattern, maps EVA's voice_id → Soniox's voice, omitted when unset so pipecat's default ("Adrian") applies.
  • Updated both "Unknown model" error messages to list soniox.
• .env.example
  • Added soniox to the STT and TTS provider enum lists.
  • Added worked example EVA_MODEL__STT_PARAMS/EVA_MODEL__TTS_PARAMS blocks for Soniox.
• docs/llm_configuration.md
  • Added a note clarifying EVA_MODEL_LIST is LLM-only and STT/TTS providers (Soniox included) go through EVA_MODEL__STT/EVA_MODEL__TTS instead — to prevent the same misconfiguration recurring.
• tests/unit/assistant/test_services.py
  • Added STT tests: service creation, optional-settings forwarding, language-hint enum conversion.
  • Added TTS tests: service creation, voice_id forwarding, default-voice fallback.
  • Full suite verified: 2028 passed, 0 failed; ruff check clean.