(fix): don't handle time-dtypes as extension arrays in from_dataframe

[ilan-gold]

This is probably the simplest way to fix the bug but I now have many questions about PandasIndexingAdapter. I initially attempted to make datetime dtypes go into ExtensionArray (because they are extension arrays) but there are so many tests with behavior that I don't understand that this seemed easier. Primarily, my issue with doing that wasn't that the tests broke in a way that was "this behavior doesn't make sense" but rather "this behavior makes more sense than what the test expects" i.e., there were a number of places where tests expect datetime dtypes to not be preserved, but they are preserved by using ExtensionArray. But I don't understand why certain things are expected or not. For example:

xarray/xarray/tests/test_variable.py

Lines 2967 to 2968 in 026aa7c

	if var.dtype.kind == "M":
	assert var.dtype == np.dtype("datetime64[ns]")

became a condition that expected the dtype be preserved and not just blanket converting everything to np.dtype("datetime64[ns]") i.e., the dtype on var remained pd.DatetimeTZDtype("s", pytz.timezone("America/New_York"))

• Closes Potential regression in Dataset.from_dataframe() not preserving timezone #9026
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[ilan-gold]

@dcherian could you review?