Eagerly populate the class dict for cudf.pandas proxy types

[shwina]

Rather than dynamically looking up class attributes (and methods), this PR makes it so that we eagerly populate the class with all known methods and attributes (by inspecting the "slow" class).

This solves a number of problems:

• it makes getattr trivially inexpensive (no dynamic __getattr__ for each attribute access)
• it ensures the same object is returned every time you do, e.g., DataFrame.max
• it makes tab completion fast because the attributes don't have to be computed each time
• it no longer exposes attributes that are specific to cuDF - for example Series.list
• it allows subclassing of proxy types to work better. For example, derived types can now call super(). to access attributes of base types

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[galipremsagar]

/okay to test

[galipremsagar]

/okay to test

[galipremsagar]

@wence- I've reverted the vast majority of the changes I made to this PR and this is now ready for review. Since I've made a couple of minor fixes to this pr I'd like to get it reviewed by someone other than me before merging.

Something to note, the drop in test pass rate is expected and are valid failures that are not in the scope of fixing in this PR.

[mroeschke]

Would be good to adjust the func typing in _maybe_wrap_result as func: Callable | None and raise an exception in is_final_type and is_intermediate_type branches if func is None

[galipremsagar]

This doesn't seem to be straight-forward, while testing this change I noticed 4k additional failures. Maybe we can revisit this in another PR.

[mroeschke]

So these attributes we are wanting the fast path to have a chance to evaluate and not automatically be _Unusable()?

[galipremsagar]

yes

[mroeschke]

Ah so in this case since it's private we won't try to operate on e.g. cudf.Index._data which is a ColumnAccessor and instead reach for cudf.Index._data which is a numpy ndarray which we can wrap.

This part of the eager population I'm not too thrilled about. These private variables are not guaranteed to be stable. It would be nice if any slow attribute was attempted to return a proxy object so we didn't have to specify attributes like this but I suppose that can be looked at in a follow up

[galipremsagar]

Yeah, there are still such cases we can address in follow-ups. At this point the scope of this PR has kept on expanding, that's why I haven't addressed additional issues in this PR.

[mroeschke]

/okay to test

[mroeschke]

My typing request and concern about having to define private attributes in our proxies can be addressed in a follow up

[galipremsagar]

/merge