feat(speech): add prompt voice input (#249)

## Summary
- add server-backed speech capabilities and transcription endpoints plus
UI settings for speech configuration
- add push-to-talk prompt voice input with microphone controls,
transcription insertion, and browser capability gating
- keep prompt controls aligned by restoring right-side nav placement and
moving the mic beside the expand control

packages/server/src/speech/service.ts

@@ -0,0 +1,91 @@
+import { z } from "zod"
+import type { Logger } from "../logger"
+import type { SettingsService } from "../settings/service"
+import type { SpeechCapabilitiesResponse, SpeechSynthesisResponse, SpeechTranscriptionResponse } from "../api-types"
+import { OpenAICompatibleSpeechProvider } from "./providers/openai-compatible"
+
+const ServerSpeechSettingsSchema = z.object({
+  speech: z
+    .object({
+      provider: z.string().optional(),
+      apiKey: z.string().optional(),
+      baseUrl: z.string().optional(),
+      sttModel: z.string().optional(),
+      ttsModel: z.string().optional(),
+      ttsVoice: z.string().optional(),
+    })
+    .optional(),
+})
+
+export interface TranscribeAudioInput {
+  audioBase64: string
+  mimeType: string
+  filename?: string
+  language?: string
+  prompt?: string
+}
+
+export interface SynthesizeSpeechInput {
+  text: string
+  format?: "mp3" | "wav" | "opus"
+}
+
+export interface SpeechProvider {
+  getCapabilities(): SpeechCapabilitiesResponse
+  transcribe(input: TranscribeAudioInput): Promise<SpeechTranscriptionResponse>
+  synthesize(input: SynthesizeSpeechInput): Promise<SpeechSynthesisResponse>
+}
+
+export interface NormalizedSpeechSettings {
+  provider: string
+  apiKey?: string
+  baseUrl?: string
+  sttModel: string
+  ttsModel: string
+  ttsVoice: string
+}
+
+const DEFAULT_PROVIDER = "openai-compatible"
+const DEFAULT_STT_MODEL = "gpt-4o-mini-transcribe"
+const DEFAULT_TTS_MODEL = "gpt-4o-mini-tts"
+const DEFAULT_TTS_VOICE = "alloy"
+export class SpeechService {
+  constructor(
+    private readonly settings: SettingsService,
+    private readonly logger: Logger,
+  ) {}
+
+  getCapabilities(): SpeechCapabilitiesResponse {
+    return this.createProvider().getCapabilities()
+  }
+
+  async transcribe(input: TranscribeAudioInput): Promise<SpeechTranscriptionResponse> {
+    return this.createProvider().transcribe(input)
+  }
+
+  async synthesize(input: SynthesizeSpeechInput): Promise<SpeechSynthesisResponse> {
+    return this.createProvider().synthesize(input)
+  }
+
+  private createProvider(): SpeechProvider {
+    const settings = this.resolveSettings()
+    return new OpenAICompatibleSpeechProvider({
+      settings,
+      logger: this.logger.child({ provider: settings.provider }),
+    })
+  }
+
+  private resolveSettings(): NormalizedSpeechSettings {
+    const parsed = ServerSpeechSettingsSchema.parse(this.settings.getOwner("config", "server") ?? {})
+    const speech = parsed.speech ?? {}
+
+    return {
+      provider: speech.provider?.trim() || DEFAULT_PROVIDER,
+      apiKey: speech.apiKey?.trim() || process.env.OPENAI_API_KEY,
+      baseUrl: speech.baseUrl?.trim() || process.env.OPENAI_BASE_URL || undefined,
+      sttModel: speech.sttModel?.trim() || DEFAULT_STT_MODEL,
+      ttsModel: speech.ttsModel?.trim() || DEFAULT_TTS_MODEL,
+      ttsVoice: speech.ttsVoice?.trim() || DEFAULT_TTS_VOICE,
+    }
+  }
+}