Support argument discovery

This is a somewhat speculative feature request.

My exploration of fsspec suggests that there's currently no standard, machine-actionable mechanism through which a backend implementation can describe the arguments (args, kwargs) that it understands.

The use-case I'm considering is for implementation-agnostic UI systems; for example the Jupyter fsspec extension. Currently, the user is expected to configure the jupyter extension by editing a configuration file, providing the necessary configuration (if any) for that fsspec backend implementation.

Seemingly, the user must discover information about how to configure the backend implementation by somehow finding and reading the backend implementation's documentation.

Fsspec already supports discovering the list of available backend implementations. If it were possible, for a specific fsspec backend implementation, to discover a list of possible configuration options, along with some metadata (input type, a brief description, whether it is required, etc) then the fsspec client (jupyter fsspec extension, in the above example) could build a UI that would allow the user to add fsspec implementations, ideally without forcing the user to discover documentation for the backend implementation.

Start by reviewing fsspec's existing mechanism for discovering available backend implementations; the issue does not name a file or test to begin with. Define the scope of a machine-actionable argument description and its metadata, then determine how the Jupyter fsspec extension would consume it. Done should include an agreed interface and validation for backend configuration discovery.