[Data] Fix WebDataset docs and encoder typing

[viiccwen]

Description

Correct the public Dataset.write_webdataset() contract so it matches the tar-based WebDataset implementation.

• Replace copied TFRecord descriptions with WebDataset sample and tar archive semantics.
• Correct output path and default extension documentation.
• Define reusable encoder aliases for a default format specifier, a sample-transform callable, or a list of those values.
• Use the aliases in both Dataset.write_webdataset() and WebDatasetDatasink.
• Add focused tests proving callable and chained-list encoders are accepted and produce the expected archive entries.

This changes documentation and static typing only; existing supported runtime inputs remain valid.

Related issues

Closes #65277.

Additional information

API documentation changes (no real behavior change).

[gemini-code-assist]

Code Review

This pull request refines the type annotations and documentation for the WebDataset writer, introducing type aliases for encoder configurations and updating the docstrings to correctly reference WebDataset tar archives instead of TFRecord files. It also adds tests to verify default, callable, and chained encoders. The review feedback suggests refining the WebDatasetEncoderConfig type alias to use List[WebDatasetEncoderSpec] instead of List[Any], and using this alias consistently in Dataset.write_webdataset instead of an inlined type union.

[gemini-code-assist]

The type alias WebDatasetEncoderConfig currently uses List[Any] for chained encoders. Since WebDatasetEncoderSpec is already defined to represent valid encoder types, we can make this type annotation much more precise and self-documenting by using List[WebDatasetEncoderSpec] instead of List[Any].

Suggested change

	WebDatasetEncoderConfig = Optional[Union[WebDatasetEncoderSpec, List[Any]]]
	WebDatasetEncoderConfig = Optional[Union[WebDatasetEncoderSpec, List[WebDatasetEncoderSpec]]]

[gemini-code-assist]

The PR description states that one of the goals is to "Use the aliases in both Dataset.write_webdataset() and WebDatasetDatasink." However, the implementation in Dataset.write_webdataset still uses an inlined (and less precise) type union.\n\nUsing the newly defined WebDatasetEncoderConfig alias here keeps the public API signature clean and consistent with WebDatasetDatasink.\n\nNote: Please also update the imports at the top of python/ray/data/dataset.py to import WebDatasetEncoderConfig from ray.data._internal.datasource.webdataset_datasink.

        encoder: "WebDatasetEncoderConfig" = True,

[viiccwen]

That's due to the style for Ray-data.