Clearer error message on unprocessable entity.

[Wauplin]

Fix #1753 by adding a better error message to the 422 unprocessable entity. I chose not to re-raise a different error but appending the message instead (easier in term of backward compatibility and still valid).

>>> from huggingface_hub import InferenceClient
>>> client = InferenceClient(model="HuggingFaceH4/zephyr-7b-alpha")
>>> output = client.conversational("Hi how are you?")
huggingface_hub.utils._errors.HfHubHTTPError: 422 Client Error: Unprocessable Entity for url: https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-alpha (Request ID: 8gb40cYT0epCW3jTQ5HUd)
Make sure 'conversational' task is supported by the model.

# python -m asyncio
>>> from huggingface_hub import AsyncInferenceClient
>>> client = AsyncInferenceClient(model="HuggingFaceH4/zephyr-7b-alpha")
>>> output = await client.conversational("Hi how are you?")
aiohttp.client_exceptions.ClientResponseError: 422, message="Unprocessable Entity. Make sure 'conversational' task is supported by the model.", url=URL('https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-alpha')

Also added some unit tests about it.

cc @jamesbraza

[jamesbraza]

Nice work! Thanks

[jamesbraza]

Suggested change

	error.args[0] + f"\nMake sure '{task}' task is supported by the model.",
	f"{error.args[0]}\nMake sure '{task}' task is supported by the model.",

I think f-str is a bit easier than concatenation. Also, why does sync client delimit with "\n" and async client delimit with ". "?

[Wauplin]

Thanks for the review! f-string makes sense yes 👍

Also, why does sync client delimit with "\n" and async client delimit with ". "?

For stylish reasons. The sync client raises a requests.HTTPError exception which print the message when serialized. I prefer the \n so that the warning is on its own line. For the async client we have a aiohttp.ClientResponseError which doesn't serialize the same way (see snippet). So I preferred to simple put ". ".
But no strong opinion on it ^^