[BUG] dataclass has no type validation so errors end up being hard to debug

[alvarobartt]

Describe the bug

This is more an improvement than a bug, because some errors end up being cryptic and hard to debug, because the client is just throwing an HTTP 500 error code while the actual traceback is in the server.

This is due to some missing validation when dataclass is used with no __post_init__ validation, instead of pydantic.BaseModel.

Stacktrace and Code to create the bug

There are more situations when this can break, but this is one of those when using TextClassificationSettings

import argilla as rg

rg.init(api_url="...", api_key="...")

rg.configure_dataset("my-dataset", workspace=rg.get_workspace(), settings=rg.TextClassificationSettings(label_schema=["A", "B"]))

Fails because it's a List[str] instead of Set[str], but the error displayed in the Python client is AttributeError: 'str' object has no attribute 'name', but the traceback cannot be followed from the Python client, just from the server.

Expected behavior

A validation layer to make sure that everything's right before calling the API, to be able to show a nice error message to the user, and better error handling, to avoid showing API exceptions straight away in the client, but a formatted version of those.

Environment:

• Argilla Version [e.g. 1.0.0]: 1.13.0 (but it's N/A)
• ElasticSearch Version [e.g. 7.10.2]: N/A
• Docker Image (optional) [e.g. argilla:v1.0.0]: argilla:v1.13.0

Additional context

This ends up with users asking for help on issues they cannot easily debug, so this would also avoid people from running into these issues or just having a simple way to solve it themselves.