Add pyannote orchestration

[EduardoPach]

What does this PR do?

This PR introduces a reusable PyannoteAI engine and adds support for PyannoteAI's new STT orchestration feature, which combines speaker diarization with transcription in a single API call.

Changes

New Engine (src/openbench/engine/pyannote_engine.py)

• Created PyannoteAIApi engine class that supports both diarization-only and diarization+transcription modes
• Added response models for API outputs:
  • PyannoteApiDiarizationOutput - diarization-only responses
  • PyannoteApiOrchestrationOutput - responses with wordLevelTranscription and turnLevelTranscription

Refactored Diarization Pipeline

• Updated PyannoteApiPipeline to use the new engine instead of duplicating API logic

New Pipelines

• PyannoteTranscriptionPipeline - Uses PyannoteAI with STT but ignores speaker attribution (for transcription-only datasets)
• PyannoteOrchestrationPipeline - Uses PyannoteAI with STT and includes speaker attribution

Pipeline Aliases

• pyannote-transcription - Transcription without speaker labels
• pyannote-orchestration - Full diarization + transcription with speaker attribution

Usage

Transcription only (no speaker labels)

openbench-cli evaluate -p pyannote-transcription -d <dataset-name> -m wer

Orchestration (with speaker labels)

openbench-cli evaluate -p pyannote-orchestration -d <dataset-name> -m wer -m cpwer -m wder

[dbrkn]

lgtm