BUG: Cannot create third-party ExtensionArrays for datetime types

[xhochy]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample: #34987

Problem description

With #33400 type checks now fasttrack on the kind of the dtype. Thus for every ExtensionDtype that is of kind M, is_datetime64tz_dtype returns True. This is then used in get_block_type which leads to the usage of DatetimeTZBlock for all of these arrays although this block only supports DatetimeArray.

pandas/pandas/core/internals/blocks.py

Lines 2685 to 2690 in 4f7d8bb

	elif is_datetime64tz_dtype(values.dtype):
	cls = DatetimeTZBlock
	elif is_interval_dtype(dtype) or is_period_dtype(dtype):
	cls = ObjectValuesExtensionBlock
	elif is_extension_array_dtype(values.dtype):
	cls = ExtensionBlock

[mroeschke]

I ran into this in #47515

Since #34987, just added some unit tests, I think this behavior should still be fixed so reopening.

[jbrockmendel]

@mroeschke will this be fixed by your is_datetime64tz_dtype PR?

[mroeschke]

I did refactor my original is_datetime64tz_dtype PR #48258 to only be a unit test since I think another one of your PRs fixed the original issue, but can open a separate PR to reintroduce the fix