[RD-567] Add TTS evals

[dbrkn]

1. Adds a speech_generation pipeline that generates audio from text prompts via whisperkit-cli tts, transcribes the output using the WhisperKitPro engine, and computes WER against the original prompt. Includes a text-only dataset, configurable TTS/transcription params, a registered alias
2. Adds a generic --pipeline-config key=value CLI flag for alias-mode overrides. ( mainly to set speakers and language for tts generation)

Sample command:

export WHISPERKIT_CLI_PATH="/path/to/whisperkit-cli"
export WHISPERKITPRO_CLI_PATH="/path/to/whisperkitpro-cli"
uv run openbench-cli evaluate \
  --pipeline whisperkit-speech-generation \
  --dataset customer-service-tts-prompts-vocalized \
  --metrics wer \
  --verbose

Sample Result: