feat(async): Improve Speech Understanding UX Through TranscriptConfig

[dmorrissey-aai]

Adjust pydantic models to allow for direct dictionary construction with validation.

What

Before:

    return aai.TranscriptionConfig(
        speech_understanding=aai.SpeechUnderstandingRequest(
            request=aai.SpeechUnderstandingFeatureRequests(
                speaker_identification=aai.SpeakerIdentificationRequest(
                    speaker_type=aai.SpeakerType.name,
                    known_values=["Michel Martin", "Peter DeCarlo"],
                )
            )
        )
    )

After:

    config = aai.TranscriptionConfig()
    config.speech_understanding = {
        "request": {
            "speaker_identification": {
                "speaker_type": "name",
                "known_values": ["Michel Martin", "Peter DeCarlo"],
            }
        }
    }
    return config

OR

    return aai.TranscriptionConfig(
        speech_understanding={
            "request": {
                "speaker_identification": {
                    "speaker_type": "name",
                    "known_values": ["Michel Martin", "Peter DeCarlo"],
                }
            }
        }
    )

Why

Broad criticism of the UX of the python sdk, with the hope of reducing new user friction.

How

By typing each attribute as Union[TypedClass, dict[str, Any]] and setting validate_assignment=True, construction and assignment of the attributes can now be made via dict passthrough while retaining actual strictness.

Existing schema, construction and assignment style still work.

Tests

New tests added proving dict pass through works and retains parity with the class-style construction