DISC: Deprecate use_inf_as_na

[dsaxton]

I think a case could be made for deprecating pd.options.mode.use_inf_as_na. Few thoughts:

• Doesn't seem necessary or particularly useful: it's easy to get the same behavior by checking for NA or infinity.
• Source of bugs: e.g., BUG: pandas.options.mode.use_inf_as_na disables NA checking for StringArray #33655, BUG: pandas 1.0 dropna error with categorical data if pd.options.mode.use_inf_as_na = True #33594. If an array has a cached mask attribute there's no guarantee that it's correct:

  import numpy as np
  import pandas as pd
  
  arr = pd.Categorical([1.0, 2.0, np.inf])
  pd.options.mode.use_inf_as_na = True
  arr.isna().any()  # False

• General maintenance cost: several null-checking functions either have to be duplicated or equipped with boolean arguments to switch their behavior (e.g., _isna and _isna_ndarraylike from /core/dtypes/missing.py, or isnaobj and isnaobj_old from /_libs/missing.pyx). These could all be cleaned up if NA had one meaning.

Thoughts?

[jbrockmendel]

xref #28392