CLN: unify fused type definitions

[jreback]

grep -r fused pandas/_libs/ --include '*.pyx'
pandas/_libs/writers.pyx:ctypedef fused pandas_string:
pandas/_libs/reshape.pyx:ctypedef fused reshape_t:
pandas/_libs/join.pyx:ctypedef fused join_t:
pandas/_libs/join.pyx:ctypedef fused asof_t:
pandas/_libs/join.pyx:ctypedef fused by_t:
pandas/_libs/algos.pyx:ctypedef fused algos_t:
pandas/_libs/algos.pyx:ctypedef fused rank_t:
pandas/_libs/algos.pyx:ctypedef fused diff_t:
pandas/_libs/algos.pyx:ctypedef fused out_t:
pandas/_libs/groupby.pyx:ctypedef fused add_t:
pandas/_libs/groupby.pyx:ctypedef fused rank_t:
pandas/_libs/groupby.pyx:ctypedef fused groupby_t:

cc @mzeitlin11 @jbrockmendel

[jreback]

maybe we need a dtypes.pyx that is shared

[mzeitlin11]

There's also the cython builtins numeric and floating we use in a few places.

It would be nice for many of these to just use numeric instead (maybe with object added in some cases) - there's a tradeoff between adding more fused types (which can improve performance by avoiding upcasting for a case like #42275), but at the cost of each new fused type increasing library size.

[jbrockmendel]

closeable?

[mroeschke]

I think we've consolidate a lot of fused types recently so closing. We can open specific issues for types can that be more fused