feat(backend,frontend,mcp): any voice or round video can be transcribed on request, media is found by its bytes

frontend/src/lib/media/transcript.svelte.ts

@@ -0,0 +1,65 @@
+import { requestTranscription } from "#lib/api/media.js";
+import { t } from "#lib/i18n/locale.svelte.js";
+import type { MessageKey } from "#lib/i18n/messages.js";
+import { toasts } from "#lib/stores/toasts.svelte.js";
+
+const REFUSALS: Record<string, MessageKey> = {
+  empty: "media.transcribeEmpty",
+  no_premium: "media.transcribeNoPremium",
+  pending: "media.transcribePending",
+  too_long: "media.transcribeTooLong",
+};
+
+export class Transcript {
+  readonly #chatId: () => number;
+  readonly #messageId: () => number;
+  readonly #initial: () => string | null;
+  fetched = $state<string | null>(null);
+  pending = $state(false);
+  shown = $state(false);
+
+  constructor(
+    chatId: () => number,
+    messageId: () => number,
+    initial: () => string | null
+  ) {
+    this.#chatId = chatId;
+    this.#messageId = messageId;
+    this.#initial = initial;
+  }
+
+  get text(): string | null {
+    return this.fetched ?? this.#initial();
+  }
+
+  get open(): boolean {
+    return this.shown && Boolean(this.text);
+  }
+
+  async toggle(): Promise<void> {
+    if (this.text) {
+      this.shown = !this.shown;
+      return;
+    }
+    if (this.pending) {
+      return;
+    }
+    this.pending = true;
+    try {
+      const outcome = await requestTranscription(
+        this.#chatId(),
+        this.#messageId()
+      );
+      if ("text" in outcome) {
+        this.fetched = outcome.text;
+        this.shown = true;
+      } else {
+        toasts.error(t(REFUSALS[outcome.error] ?? "media.transcribeFailed"));
+      }
+    } catch {
+      toasts.error(t("media.transcribeFailed"));
+    } finally {
+      this.pending = false;
+    }
+  }
+}