Fix: Allow custom client to be passed to ChatDatabricks (Issue #128)

[jayshrivastava0]

This commit addresses issue #128 by modifying the __init__ method of the ChatDatabricks class to allow a custom client to be passed in via the constructor.

Previously, any client passed in the constructor would be overwritten by the default client. With this change:

• If a client is provided in kwargs, that client instance is used.
• Otherwise, the client is initialized using get_deployment_client(self.target_uri).

A new unit test, test_chat_model_with_custom_client, has been added to integrations/langchain/tests/unit_tests/test_chat_models.py to verify this functionality. The test ensures that the provided custom client is correctly assigned and its methods are called when the ChatDatabricks instance makes predictions.

[aravind-segu]

@jayshrivastava0 what is the usecase for passing in the client directly? What kind of clients are you passing in here?

[jayshrivastava0]

Hey @aravind-segu, great question. We enabled this to give developers more flexibility.

The two main use cases are:

1. Testing: It allows you to inject a mock client for unit tests, so you can verify your logic without making actual network calls.
2. Custom Configuration: You can pass in a client that's already configured with specific enterprise needs, like custom authentication, proxy settings, or advanced retry logic.

The client just needs to be an object that behaves like MLflow's BaseDeploymentClient.