Add /v1/audio/transcriptions endpoint for Whisper support

[elimisteve]

Summary

maple-proxy should expose an OpenAI-compatible /v1/audio/transcriptions endpoint, enabling audio transcription via whisper-large-v3 through the proxy.

Blocked by

This is blocked on the opensecret Rust SDK adding transcription support:
OpenSecretCloud/opensecret#140

Once OpenSecretClient has a create_transcription() method, maple-proxy needs a new route that:

1. Accepts multipart/form-data with an audio file, model name, and optional params (language, response_format)
2. Calls the new OpenSecretClient::create_transcription() method
3. Returns the transcription response

Context

The Maple backend already lists whisper-large-v3 as an available model, but there's no way to use it through the proxy today.