TST: Use numeric dtype fixtures for testing

[jschendel]

Some new numeric dtype fixtures were implemented in #21432:

pandas/pandas/conftest.py

Lines 183 to 243 in 9e982e1

	@pytest.fixture(params=["float32", "float64"])
	def float_dtype(request):
	"""
	Parameterized fixture for float dtypes.
	* float32
	* float64
	"""
	return request.param
	UNSIGNED_INT_DTYPES = ["uint8", "uint16", "uint32", "uint64"]
	SIGNED_INT_DTYPES = ["int8", "int16", "int32", "int64"]
	ALL_INT_DTYPES = UNSIGNED_INT_DTYPES + SIGNED_INT_DTYPES
	@pytest.fixture(params=SIGNED_INT_DTYPES)
	def sint_dtype(request):
	"""
	Parameterized fixture for signed integer dtypes.
	* int8
	* int16
	* int32
	* int64
	"""
	return request.param
	@pytest.fixture(params=UNSIGNED_INT_DTYPES)
	def uint_dtype(request):
	"""
	Parameterized fixture for unsigned integer dtypes.
	* uint8
	* uint16
	* uint32
	* uint64
	"""
	return request.param
	@pytest.fixture(params=ALL_INT_DTYPES)
	def any_int_dtype(request):
	"""
	Parameterized fixture for any integer dtypes.
	* int8
	* uint8
	* int16
	* uint16
	* int32
	* uint32
	* int64
	* uint64
	"""
	return request.param

It would be nice to use these fixtures where appropriate in the existing tests, e.g.

pandas/pandas/tests/indexes/datetimes/test_construction.py

Lines 527 to 528 in 9e982e1

	@pytest.mark.parametrize('dtype', [np.int64, np.int32, np.int16, np.int8])
	def test_dti_constructor_small_int(self, dtype):

Additional comments:

• These changes can be split across multiple PR's so no need to do all changes at once
  • Not how many instances of this will need to be replaced, so this will need to be investigated
• This might involve expanding the dtype coverage for some tests
  • The example test shown above could also be tested over uint dtypes
• Can probably drop float16 dtypes from tests since it's barely supported
  • See BUG: Fix Series.nlargest for integer boundary values #21432 (comment)
• Might be necessary to create additional fixtures
  • Perhaps a fixture for the combination of float + signed integer + unsigned integer dtypes

[gfyoung]

They have propagated since this issue was opened. Just need to make sure we use them in new PR's.