refactor(server): extract SpeechService and defer init until after listen

Speech runtime was leaking 8+ individual resolvers into bootstrap and
blocking server startup for ~3s with synchronous Sherpa native model
loading. Refactor into a self-contained SpeechService that owns its
full lifecycle (init, download, monitor, cleanup) and defer start()
until after httpServer.listen() so native loading doesn't block
accepting connections. Server startup drops from ~3.2s to ~0.5s.

Also removes unused client-triggered speech model download/list path
(CLI commands, session handlers, message types) — the service manages
its own models.

packages/server/src/client/daemon-client.ts

@@ -39,8 +39,6 @@ import type {
   ListCommandsResponse,
   ListProviderModelsResponseMessage,
   ListAvailableProvidersResponse,
-  SpeechModelsListResponse,
-  SpeechModelsDownloadResponse,
   ListTerminalsResponse,
   CreateTerminalResponse,
   SubscribeTerminalResponse,
@@ -217,8 +215,6 @@ type FileExplorerPayload = FileExplorerResponse["payload"];
 type FileDownloadTokenPayload = FileDownloadTokenResponse["payload"];
 type ListProviderModelsPayload = ListProviderModelsResponseMessage["payload"];
 type ListAvailableProvidersPayload = ListAvailableProvidersResponse["payload"];
-type SpeechModelsListPayload = SpeechModelsListResponse["payload"];
-type SpeechModelsDownloadPayload = SpeechModelsDownloadResponse["payload"];
 type ListCommandsPayload = ListCommandsResponse["payload"];
 type ListCommandsDraftConfig = Pick<
   AgentSessionConfig,
@@ -2487,32 +2483,6 @@ export class DaemonClient {
     });
   }
 
-  async listSpeechModels(requestId?: string): Promise<SpeechModelsListPayload> {
-    return this.sendCorrelatedSessionRequest({
-      requestId,
-      message: {
-        type: "speech_models_list_request",
-      },
-      responseType: "speech_models_list_response",
-      timeout: 30000,
-    });
-  }
-
-  async downloadSpeechModels(options?: {
-    modelIds?: string[];
-    requestId?: string;
-  }): Promise<SpeechModelsDownloadPayload> {
-    return this.sendCorrelatedSessionRequest({
-      requestId: options?.requestId,
-      message: {
-        type: "speech_models_download_request",
-        modelIds: options?.modelIds,
-      },
-      responseType: "speech_models_download_response",
-      timeout: 30 * 60 * 1000,
-    });
-  }
-
   async listCommands(agentId: string, requestId?: string): Promise<ListCommandsPayload>;
   async listCommands(agentId: string, options?: ListCommandsOptions): Promise<ListCommandsPayload>;
   async listCommands(