fix azure by removing model checking

[PCSwingle]

Fixes #489

Pull Request Checklist

• Documentation has been updated, or this change doesn't require that

[biobootloader]

If the user sets the model name as gpt-35-turbo (instead of gpt-3.5-turbo, like in the issue that prompted this), then will we not have cost estimates for them?

[PCSwingle]

If the user sets the model name as gpt-35-turbo (instead of gpt-3.5-turbo, like in the issue that prompted this), then will we not have cost estimates for them?

We won't but there's no way to do that; you name the azure model when you set it up, so it could be anything.

[jakethekoenig]

Two things that might make sense to do (not as part of this pr):

1. Let the user set the costs in the config. Sort of like how they can set maximum_context.
2. Guess what prices we should use by embedding their model name and seeing what model it's closest to.

[biobootloader]

how about let the user set a base model to use for cost estimate as well as the model name to call?

so there'd be an extra variable like azure_deployment_true_model_name or something. If they don't set it then they'd get a warning when using azure that we don't know model limits or costs unless they set that

[PCSwingle]

I think this adds too much complexity for no good reason; if anything, we would want to add a config option to set the input and output token prices (but I also don't think we should do that since it's too complicated for so little gain).