CLN: revisit build warnings in cython templates

[ghost]

• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Retrying dropped portions from #27157, which amend cython templates, to deal with special-cases based on type. In this Pr, compiler warnings about x==x being useless in the case of ints, but not floats (since nan!=nan).

In #27157 (comment) these were rejected because there wasn't agreement about how this should done and a separate discussion was requested. Annoyingly, these changes agree with what already exists in these files (in multiple places):

pandas/pandas/_libs/groupby_helper.pxi.in

Lines 212 to 216 in 355e322

	{{if name == 'int64'}}
	mask = (masked_vals == {{nan_val}}).astype(np.uint8)
	{{else}}
	mask = np.isnan(masked_vals).astype(np.uint8)
	{{endif}}

But the file is already inconsistent in how these type checks are done:

pandas/pandas/_libs/groupby_helper.pxi.in

Line 391 in 355e322

if groupby_t is int64_t:

Making a decision here also required to get rid of the warnings mentioned in #27169, in that case fused types which mix signed/unsigned trigger warning when the code later includes comparisons with a signed quantity. So there too, type checks in the template are the obvious fix.

cc @WillAyd, @jreback

[WillAyd]

I'm not really sure this is an improvement as it just adds more layers to the templating where we already have a weird mix between templates and fused types. Is there a way to just do this all using fused types perhaps?

[jreback]

actually i think these changes are ok, though can you have a look and see how consistent we are with these idioms across cython? is there a way to centralize these templates (easily)? e.g. have a is_null type of template

[ghost]

is there a way to centralize these templates (easily)?

maybe, I'll try it. But then, what would to do with the signed/unsigned compares in #27157?

[jreback]

ok this is positive progress towards reducing warnings.