We forbid to add nan values to the payload in local mode, since the server does not allow to have them.
An attempt to send a request with nan values fails.

However, as of pydantic 2.7, nan values are converted to null, and such requests do not fail.
pydantic/pydantic#8440
pydantic/pydantic-core#1251

Even if we'd like to support nan values in local mode when pydantic>=2.7 is installed, there is no easy way to convert nan values to null in local mode.

This PR suggests to convert nan and inf to null and aligns local mode with pydantic>=2.7

This PR suggests to disable such tests with the remote mode, we'll have a discrepancy then: local mode won't support nan values, server mode won't support them with pydantic<2.7, and will support them with pydantic>=2.7.

One of the alternatives is to convert all the nans and infs to None manually, e.g. with

def convert_nan_inf_to_null(obj):
    if isinstance(obj, dict):
        return {k: convert_nan_inf_to_null(v) for k, v in obj.items()}
    elif isinstance(obj, list):
        return [convert_nan_inf_to_null(v) for v in obj]
    elif isinstance(obj, float):
        if np.isnan(obj) or np.isinf(obj):
            return None
    return obj

Another alternative which I also considered is to extend all the models, e.g. from

class VectorParams(BaseModel, extra="forbid"):
    ...

to

class VectorParams(BaseModel, extra="forbid", allow_inf_nan=False):
    ...

However, it won't work since we use construct and validation does not happen