Add mechanic to override default values for generation during model.predict()

[justinxzhao]

Sample code:

model = LudwigModel(config=config, logging_level=logging.INFO)
results = model.train(training_set=dataset)
model.predict(dataset=dataset[:1], generation_config={"max_new_tokens": 513})

I've added additional logging statements to make it easier for testers/users to verify behavior.

[github-actions]

Unit Test Results

  6 files  ±0    6 suites  ±0   1h 7m 54s ⏱️ + 2m 41s
34 tests ±0  29 ✔️ ±0    5 💤 ±0  0 ❌ ±0 
88 runs  ±0  72 ✔️ ±0  16 💤 ±0  0 ❌ ±0 

Results for commit 1b6ee20. ± Comparison against base commit 92d07f6.

[arnavgarg1]

LGTM, thanks for adding this!

As discussed offline, if the logging gets noisy, we can just change the logging.info to logging.debug