feat: LLM - Made tuning asynchronous when tuning becomes GA

[copybara-service]

feat: LLM - Made tuning asynchronous when tuning becomes GA

Previously, tune_model waited for the tuning is complete, then modified the model in-place.
This behavior will change in the future GA (non-preview) classes:

In the future, tune_model will become asynchronous: It will start tuning job and return a job object immediately without waiting. This will allow the user to do other work while the model is being tuned. This will also allow the user to perform multiple tuning jobs in parallel.

Future breaking change: The model will no longer be updated in-place, so the user will need to get the tuned model from the job object.

To make the transition easier and avoid breaking changes, the .tune_model(...) method will start returning the job object even in preview classes (although it will still wait for the job completion and update the model in-place too). This makes it possible to start writing future-proof code immediately.

Usage:

tuning_job = model.tune_model(...)  # Returns tuning job. In preview: Waits for the tuning job to finish
tuned_model = tuning_job.get_tuned_model()  # Returns tuned model after waiting for the tuning job to finish.