Documentation-related fixes to the python client

[abidlabs]

This PR:

• Adds docstrings for the Client class and methods, clarifies that callbacks are executed sequentially
• Adds usage information for a given API. For example, for this Space (https://abidlabs-stt-or-tts.hf.space/), getting the usage info looks like this:

client = grc.Client(space="abidlabs/stt_or_tts")
client.view_api()

which produces:

Client.predict() Usage Info
---------------------------
Named endpoints: 2
 - predict(input, api_name="predict") -> audio
    Parameters:
     - [Textbox] input: str (value)
    Returns:
     - [Audio] audio: str (filepath or URL)
 - predict(input, api_name="predict_1") -> output
    Parameters:
     - [Audio] input: str (filepath or URL)
    Returns:
     - [Textbox] output: str (value)

Unnamed endpoints: 1
 - predict(parameter_0, fn_index=2) -> input,audio
    Parameters:
     - [Dataset] parameter_0: str (value)
    Returns:
     - [Textbox] input: str (value)
     - [Audio] audio: str (filepath or URL)

You can also get the API in a machine-friendly way by simply doing client.view_api(return_info=True), so this closes #2364

In order to get this info programmatically, I expanded the set of Serializers.

Related: #3565

[gradio-pr-bot]

All the demos for this PR have been deployed at https://huggingface.co/spaces/gradio-pr-deploys/pr-3663-all-demos

[abidlabs]

Although still need to add tests and fix old tests, I'd appreciate a review for feedback on this documentation format and approach for getting the documentation.

cc @aliabd -- I was able to get all of the documentation entirely from the backend. I think we can remove the documentation from the frontend and use this instead.

[abidlabs]

Synced with @freddyaboulton, I'll make some changes to this PR before opening it for reivew

[abidlabs]

@freddyaboulton after our discussion, made the following changes:

• Added usage info about each component to the config
• For backwards compatibility, the client looks for the usage information in the config. If not, falls back to what's provided by the serializer classes

For now, the only component that overrides the serializer info is the Slider(), to add information about its own maximum and minimum. In the future, we could extend this to other components, although imo it's better not to have too many component-specific usage info if the serializers can be sufficient.

Adding some tests and then will open this up for review

[abidlabs]

Opening this up for review

(Note that GitHub actions is currently down, but ci.sh can be run locally)

[freddyaboulton]

Solid PR @abidlabs ! Implementation looks good to me. Left some comments but nothing major. Thank you!!

[freddyaboulton]

This looks so slick for the stable diffusion space BTW! 🔥

[freddyaboulton]

@abidlabs Thanks for making the changes! This looks great. I guess we need to release to PyPi to fix the CI? Feel free to merge once CI is green - landing soonish.

[abidlabs]

Thanks @freddyaboulton for the great suggestions! Yup still doing some more tweaks to get the machine readable part working, but I'll release to pypi once everything is working

[abidlabs]

All righty, did some more cleanup, now we have a machine-parsable info, everything is passing, and docs looks pretty nice imo.

Will release a new version of the client to pypi, and then merge once CI passes.

[abidlabs]

Lots of little annoying fixes, but we should be there soon (cc @aliabd)

[abidlabs]

Thanks for the review @freddyaboulton, finally in!