fix: strip duplicated resolution frame validation

[nikbpetrov]

fixes #206

also happens to fix a dormant deviation for infer: original infer refactor stripped infer's astype(dtype=constants.RESOLUTION_FILE_COLUMN_DTYPE) when building the resolution frame. Is now corrected and matches manifold/metaculus.

Let me know if I should run parity tests for all 3 sources - I will assume not.

[nikbpetrov]

@houtanb now also fixed polymarket as it was duplicating validation too but I'd also like to ask your advice on whether I should just refactor the semantics of resolution file building altogether - I don't like it anymore.

Some sources have both _build_resolution_file as well as _finalize_resolution_df, while others don't (seems like a data/market source split). I don't see the point in that - we don't use/centralize these abstractions. I'd suggest we stick to only _build_resolution_df that each source (optionally) has and all the stuff inside _finalize_resolution_df (for infer/manifold/metaculus) just gets moved to _build_resolution_df (which is the rename of _build_resolution_file.

What do you think?

[houtanb]

Should this return pd.DataFrame now that validation has moved to UpdateResult.__post_init__? Currently DataFrame[ResolutionFrame] implies this helper returns a validated schema.

[houtanb]

Same question for:

    def _build_resolution_file(
        self,
        question: dict,
        resolved: bool,
        existing_df: DataFrame[ResolutionFrame] | None = None,
    ) -> DataFrame[ResolutionFrame]:

[nikbpetrov]

I want to keep the annotation for readability reasons. I appreciate the implicitness of the annotation but it's not decorated with @pa.check_types so validation does not apply.

[nikbpetrov]

#215