feat: add filter and timestamp splits

[ivanmkc]

Changes to training splits:

• Defers default splits to the server instead of setting in client
• Raises error when multiple split types passed
• Adds filter splits (b/172365904) to image, text and video
• Adds timestamp split (b/172368070) to tabular
• Added tests for all split types

Future:

• next PR will change forecasting training jobs to extend tabular; therefore, adding timestamp and predefined split to forecasting

Continued from: #549

[ivanmkc]

Moved default setting responsibility to server-side instead of in multiple places in the client.

[ivanmkc]

Switch to None, hence deferring to server.

[ivanmkc]

This is a weird thing we have to do for AutoMLVideo, pending discussion with Video team.

[ivanmkc]

Removed since we test splits separately later.

[sasha-gitg]

Thanks Ivan! LGTM and left a few comments.

[sasha-gitg]

Suggested change

	Assigns input data to training, validation, and test sets
	If using filter splits all of ``training_filter_split``, ``validation_filter_split`` and
	``test_filter_split`` must be provided.
	Assigns input data to training, validation, and test sets

Please apply this to class docstrings that support filter splits.

[ivanmkc]

Done.

[sasha-gitg]

Suggested change

	Assigns input data to training, validation, and test sets
	If using timestamp split please provide ```timestamp_split_column_name```
	Any of ``training_fraction_split``, ``validation_fraction_split`` and
	``test_fraction_split`` may optionally be provided.
	Assigns input data to training, validation, and test sets.

Please apply this to class docstrings that support timestamp splits.

[ivanmkc]

Will lint and merge.

[ivanmkc]

Merged without passing 'Sample Lint' check as I didn't modify sample code.