feat: auto-select STT engine (Azure → Google Cloud → Whisper) and show provider in results

Removes user-facing engine/model/key/beam controls. The server now picks
the best available engine automatically:
1. Microsoft Azure Speech — short clips (≤1MB, no diarization, audio/*)
2. Google Cloud Speech v2 — long audio, diarization, all languages
3. OpenAI Whisper GPU — local fallback

Results display which provider was used (e.g. "Transcribed with Google
Cloud Speech") via transcript-engine-badge and traceMeta.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Do Better Tools v1/design/inter.css

@@ -0,0 +1,6 @@
+/* Inter — pulled from Google Fonts.
+   The tools.css :root specifies "Inter, ui-sans-serif, system-ui, …".
+   No webfont @font-face is shipped with the codebase, so we substitute with
+   the official Google Fonts CDN copy at the standard weights used (400/500/
+   600/700/800/900). If you have an in-house woff2, swap this file. */
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800;900&display=swap');