[OpenAI] Audio Outputs

[hudson-ai]

This PR adds support for gen_audio (potentially not the final public API) for OpenAI audio-enabled models.

Note that this involves a pretty significant refactor of the initial implementation of "State" objects introduced a PR or two ago and centralizes handling of different input/output formats in Client classes. They may need to be renamed now that they do more heavy lifting. But in all, I think this is far more flexible and easy to extend.

@nopdive this also paved the way to re-introduce token probabilities in a pretty full capacity -- will continue working on that in the coming days.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 53.74332% with 173 lines in your changes missing coverage. Please review.

Project coverage is 48.46%. Comparing base (9fe8b26) to head (8f7b94d).
Report is 5 commits behind head on main.

Files with missing lines	Patch %	Lines
guidance/models/_openai.py	31.69%	125 Missing ⚠️
guidance/models/_engine/_client.py	58.00%	21 Missing ⚠️
guidance/models/_base/_client.py	75.67%	9 Missing ⚠️
guidance/_ast.py	84.37%	5 Missing ⚠️
guidance/library/_image.py	28.57%	5 Missing ⚠️
guidance/models/_base/_model.py	91.17%	3 Missing ⚠️
guidance/models/_base/_state.py	87.50%	2 Missing ⚠️
guidance/models/transformers/_model.py	60.00%	2 Missing ⚠️
guidance/library/_audio.py	66.66%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1136      +/-   ##
==========================================
+ Coverage   43.44%   48.46%   +5.01%     
==========================================
  Files          72       78       +6     
  Lines        5849     5586     -263     
==========================================
+ Hits         2541     2707     +166     
+ Misses       3308     2879     -429     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Harsha-Nori]

Apologies if I missed it in the PR, but @hudson-ai do you have a small code snippet I can run to play with the feature?

[hudson-ai]

from guidance import *

lm = models.OpenAI("gpt-4o-audio-preview")
with system():
    lm += "Talk like a pirate."
with user():
    lm += "What is the capital of France?"
with assistant():
    lm += gen_audio()

@Harsha-Nori give this a shot! Not without bugs at the moment 😜