Support advanced prompt voice agent configuration

[JianW (v1212)]

Summary

• Builds on the previously merged prompt voice agent support by adding advanced, editable prompt voice configuration in azure.yaml.
• Aligns the azd authoring shape with the current sample expectations and Vienna prompt voice implementation surface: structured inputs, audio input/output tuning, output modalities, tools, and avatar configuration.
• Keeps existing simple prompt voice projects working by layering advanced overrides on top of the current defaults for model, instructions, voice, store, and audio pipeline.

Scope and compatibility

• This PR is a public-preview follow-up for kind: prompt-voice; it does not change hosted/container agent deployment behavior.
• Existing prompt voice YAML remains valid. Minimal managed/BYOM agents can still use only modelType, model.id, instructions, voice, and store.
• Advanced fields are optional and map to the prompt voice service contract where supported. Missing audio fields keep azd defaults.
• Other agent paths were reviewed for isolation: hosted/container mapping, code/container deploy, provisioning config, and non-voice service round-tripping remain unchanged.

Field coverage and source of truth

• audio, outputModalities, structuredInputs, tools, avatar, and store follow the Vienna prompt voice contract/tests and current prompt voice samples.
• azd validates stable local constraints that Vienna also validates, including supported audio formats (audio/pcm, audio/pcmu, audio/pcma), speed range, server VAD threshold range, and required avatar/voice fields.
• Extensible service areas remain intentionally light: tool objects and avatar extension fields are passed through to prompt voice service validation so azd does not block future Vienna/Voice Live additions.
• structuredInputs.defaultValue uses the azure.yaml camelCase authoring style and is converted to the service wire field default_value during deployment.

Implementation

• Extends prompt voice authoring models and REST payload models for advanced voice config.
• Updates mapping so audio, structuredInputs, outputModalities, tools, and avatar flow from YAML to the create request.
• Adds validation for local constraints such as audio format, turn detection ranges, output speed, and blank modalities.
• Updates the extension JSON schema and README with the advanced azure.yaml shape, known values, and pass-through boundaries.

Validation

• go test ./... -count=1
• go build ./...
• cspell lint 'extensions/**/*.go' 'extensions/**/*.md' --config ./.vscode/cspell.yaml --no-progress
• Live managed prompt voice deploy in France Central with advanced YAML: structured inputs, server VAD, transcription, Azure voice style, system tool, avatar, store; control-plane GET assertions passed.
• Live BYOM realtime prompt voice deploy in France Central with advanced YAML: self-deployed realtime model, semantic VAD, PCMU input, audio+text output modalities, function tool, structured inputs; control-plane GET assertions passed.
• Negative live smoke: audio/opus is rejected locally before deploy with a supported-format validation error.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
20 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.