Add speech-to-text definitions for 8 providers

[ajot]

Add speech-to-text support across providers that offer STT endpoints.

Phase 1 — URL-based STT (no file upload):

• AssemblyAI (polling pattern)
• Groq, Together, Fireworks, Mistral (sync/multipart)

Requires:

• multipart/form-data support in proxy.py (currently only handles JSON bodies)
• New url UI param type for audio URL input

Phase 2 — File upload:

• OpenAI Whisper, DeepInfra, SambaNova
• Needs a file upload UI component and binary body handling

A design plan exists in docs/plans/2026-02-23-speech-to-text-plan.md.