Create a low-level interface for websocket transcription

[jcdyer]

• Add low-level websocket handle to transcription api
• fmt

[bd-g]

Feels much cleaner.

Accidentally clicked approve - meant to just comment, will review more in depth tomorrow

[bd-g]

The code itself looks clean, but something here is breaking the simple_stream example - I can get it to run successfully on main, but not this branch.

With that said, we also need to update examples/README.md. The filename is incorrect, we should suggest exporting an API key, and we should update the example names.

---

Running Examples

Setting Env Vars

export FILENAME=./examples/bueller.wav
# If using the hosted Deepgram API
export DEEPGRAM_API_KEY=<your_api_key_here>

Running the examples

cargo run --example prerecorded_from_file
cargo run --example callback
cargo run --example make_prerecorded_request_builder
cargo run --example prerecorded_from_url
cargo run --example simple_stream
cargo run --example microphone_stream
cargo run --example text_to_speech_to_file
cargo run --example text_to_speech_to_stream

[DamienDeepgram]

LGTM

[bd-g]

Paired with Cliff to review, committed requested changes.

One notable TODO that remains: the stream loop assumes a given future will interoperate with tokio::select_biased cleanly. When testing with futures::channels streams, the futures::select_biased does not properly bias towards the websocket channel, and can get stuck in an infinite loop receiving None messages from the closed audio channel.