DEPR: Series.round and Series.clip on non-numeric and non-object dtypes

[rhshadrach]

Ref: #63444

Prior to the PR linked to the above issue, Series.round would raise on only object dtype and be a no-op on other dtypes that did not support rounding. The behavior is also the same for Series.clip.

ser = pd.Series(pd.Categorical([1.234], categories=[1.234]))

print(ser.abs())
# TypeError: Object with dtype category cannot perform the numpy op absolute

print(ser.round())
# 0    1.234
# dtype: category
# Categories (1, float64): [1.234]

print(ser.clip())
# 0    1.234
# dtype: category
# Categories (1, float64): [1.234]

ser = pd.Series(["a"], dtype=str)

print(ser.abs())
# pyarrow.lib.ArrowNotImplementedError: Function 'abs_checked' has no kernel matching input types (large_string)

print(ser.round())
# 0    a
# dtype: str

print(ser.clip())
# 0    a
# dtype: str

I think we should deprecate Series.round and Series.clip from being a no-op. The corresponding DataFrame methods are to still be no-ops on such columns.

[Bhumi472]

Thanks for raising this. I agree that having Series.round and Series.clip silently act as no-ops on non-numeric dtypes can be confusing, especially since Series.abs already raises in these cases.
From a user perspective, the current behavior makes it harder to detect mistakes early, as operations appear to succeed but have no effect. Deprecating this behavior for Series seems reasonable and more consistent with the principle of explicit failure.
Keeping the DataFrame methods as no-ops also makes sense for backward compatibility and mixed-dtype workflows. Overall, this looks like a good step toward clearer and more predictable behavior.

[jbrockmendel]

I'm fine with deprecating this, but think for consistency the DataFrame cases need to be deprecated too.