feat: expose speaker embeddings and subsegments in DiarizeResult

[smm-h]

Summary

The diarize() function already computes speaker embeddings and subsegments via extract_embeddings(), but these are discarded before building the DiarizeResult. This change simply preserves them on the result object by adding two new optional fields.

Changes

• utils.py: Added embeddings: Any = None and subsegments: list[SubSegment] | None = None fields to DiarizeResult. Added model_config = ConfigDict(arbitrary_types_allowed=True) to support numpy arrays in the Pydantic model.
• __init__.py: Pass embeddings and subsegments to the DiarizeResult constructor in diarize().

Motivation

Use case: cross-recording speaker clustering and identification. When processing multiple audio files, having access to the raw speaker embeddings allows users to cluster or match speakers across recordings -- something that is not possible with just the segment labels.

Notes

• Both fields default to None, so the change is fully backward-compatible.
• No performance impact -- this just stores a reference to already-computed data instead of discarding it.