feat(agent): add server→client AgentResponse enum

  Adds the full server-emitted JSON event surface for the Voice Agent
  WebSocket (16 typed variants plus an Unknown(serde_json::Value)
  catch-all for forward compatibility).

  Variants mirror AgentV1*Event in
  api/specs/asyncapi/schemas/agent/schemas.agent.v1.yml:
    - Welcome (request_id)
    - SettingsApplied
    - ConversationText (role, content, +languages/languages_hinted on
      user-role messages with flux-general-multi)
    - UserStartedSpeaking
    - AgentThinking (content)
    - FunctionCallRequest (functions[]: id, name, arguments, client_side,
      thought_signature?)
    - AgentStartedSpeaking (total/tts/ttt latency floats; experimental)
    - AgentAudioDone
    - Error / Warning (description, code)
    - History (reuses HistoryMessage — same wire shape as
      agent.context.messages[])
    - PromptUpdated, SpeakUpdated, ThinkUpdated
    - InjectionRefused (message)
    - FunctionCallResponse (reuses the bidirectional
      FunctionCallResponseMessage from messages.rs)

  Wire dispatch follows the same #[serde(untagged)] + per-variant marker
  enum pattern as ClientMessage. Unknown is the tail variant so any
  unrecognized server message round-trips its raw JSON instead of
  failing deserialization — matches the FluxResponse forward-compat
  strategy.

  Audio frames sent by the server are binary WebSocket frames and are
  not part of this enum; they'll be delivered out-of-band by the
  connection layer when that surface lands.

  21 new round-trip serde tests, 125 passing across the agent module.

Author: GregHolmes

src/agent/mod.rs

@@ -20,17 +20,25 @@
 //!   (`UpdateSpeak`, `UpdateThink`, `UpdatePrompt`, `InjectUserMessage`,
 //!   `InjectAgentMessage`, `FunctionCallResponse`, `KeepAlive`) plus the
 //!   `ClientMessage` discriminated union over all eight.
+//! - [`response`] — every server-emitted JSON event (`Welcome`,
+//!   `SettingsApplied`, `ConversationText`, `UserStartedSpeaking`,
+//!   `AgentThinking`, `FunctionCallRequest`, `AgentStartedSpeaking`,
+//!   `AgentAudioDone`, `Error`, `Warning`, `History`, `PromptUpdated`,
+//!   `SpeakUpdated`, `ThinkUpdated`, `InjectionRefused`,
+//!   `FunctionCallResponse`) plus an `Unknown` catch-all for forward
+//!   compatibility, all unified under the [`response::AgentResponse`] enum.
 //!
 //! Wire format matches the AsyncAPI schemas in `deepgram-docs` under
-//! `api/specs/asyncapi/schemas/agent/`. The server-to-client event
-//! enum and the WebSocket connection helpers land in subsequent commits.
+//! `api/specs/asyncapi/schemas/agent/`. The WebSocket connection helpers
+//! (builder, handle, stream) land in subsequent commits.
 
 pub mod audio;
 pub mod aws_credentials;
 pub mod endpoint;
 pub mod history;
 pub mod listen;
 pub mod messages;
+pub mod response;
 pub mod settings;
 pub mod speak;
 pub mod think;
@@ -54,6 +62,15 @@ pub use messages::{
     InjectUserMessageType, KeepAliveMessage, KeepAliveType, UpdatePromptMessage, UpdatePromptType,
     UpdateSpeakMessage, UpdateSpeakType, UpdateThinkMessage, UpdateThinkType,
 };
+pub use response::{
+    AgentAudioDoneEvent, AgentAudioDoneType, AgentFunctionCall, AgentResponse,
+    AgentStartedSpeakingEvent, AgentStartedSpeakingType, AgentThinkingEvent, AgentThinkingType,
+    ConversationTextEvent, ConversationTextType, ErrorEvent, ErrorType, FunctionCallRequestEvent,
+    FunctionCallRequestType, InjectionRefusedEvent, InjectionRefusedType, PromptUpdatedEvent,
+    PromptUpdatedType, SettingsAppliedEvent, SettingsAppliedType, SpeakUpdatedEvent,
+    SpeakUpdatedType, ThinkUpdatedEvent, ThinkUpdatedType, UserStartedSpeakingEvent,
+    UserStartedSpeakingType, WarningEvent, WarningType, WelcomeEvent, WelcomeType,
+};
 pub use settings::{
     AgentConfig, AgentContext, InlineAgentConfig, SettingsFlags, SettingsMessage,
     SettingsMessageType,