Add SmallestAI Pulse STT and Lightning TTS to pipecat cascade pipeline

[weiz9]

Summary

• Wires pipecat's SmallestSTTService (Pulse) and SmallestTTSService (Lightning v3.1 / v3.1 Pro) into create_stt_service/create_tts_service, so the CASCADE pipeline (STT -> LLM -> TTS) can use Smallest as a provider alongside the existing SmallestAI Hydra S2S integration.
• Requires pipecat-ai>=1.4.0 (already the pinned minimum in pyproject.toml).
• Bumped simulation_version (2.0.1 -> 2.0.2) since this touches src/eva/assistant/.
• Added smallest to the STT/TTS provider enums in .env.example.

Test plan

• pytest tests/unit/assistant/test_services.py -v — new smallest STT/TTS unit tests pass, including the default-voice-preservation edge case (voice_id omitted -> pipecat's per-model default voice applies).
• pytest tests/unit -q — full unit suite passes (1846 passed).
• ruff check / ruff format --check clean; mypy shows no new errors beyond the pre-existing api_key: Any | None pattern shared by every other provider in this file.
• End-to-end EVA_DEBUG=true eva --domain airline with stt=smallest (pulse), tts=smallest (lightning_v3.1_pro), llm=gpt-5.4: 3/3 conversations succeeded. Verified valid non-silent 24kHz audio, clean multi-turn transcripts, zero error frames from SmallestSTTService/SmallestTTSService in pipecat logs, and reasonable STT WER (~9%).