feat(assemblyai): add speaker diarization support

[dlange-aai]

Summary

• Extract speaker_label from Turn responses and populate speaker_id on SpeechData
• Declare diarization capability when speaker_labels is enabled, allowing use with MultiSpeakerAdapter
• Map UNKNOWN speaker labels (short utterances <1s) to None

Test plan

• Tested locally with u3-rt-pro model, speaker_labels=True and confirmed speaker IDs ("A", "B") appear in transcription events
• Verified speaker=None before change, speaker=A/B after

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

David Lange seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 2 additional findings.

[longcw]

looks good to me!

[dlange-aai]

Thanks! Is this blocked from merging due to the CLA agreement? I signed but it's having an issue reconciling to my Github account.