Handle np.datetime64 and other numpy dtypes with odd subclasses/constructors

[sneakers-the-rat]

Fix: #46

np.datetime64 isn't a dtype class like other dtypes, it produces a different type:

>>> arr = np.array([np.datetime64(datetime.now())], dtype=np.datetime64)
>>> arr.dtype
dtype('<M8[us]')
>>> type(arr.dtype)
numpy.dtypes.DateTime64DType

But the type attribute is unambiguous:

>>> arr.dtype.type
numpy.datetime64

This is similar to numpy string dtypes.

Fix is simple and generic - just check the type attribute if it's present.

opening first with failing test, then with fix

[coveralls]

Pull Request Test Coverage Report for Build 13777381869

Details

• 1 of 1 (100.0%) changed or added relevant line in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 98.302%

---

Totals
Change from base Build 12941001383:	0.0%
Covered Lines:	1505
Relevant Lines:	1531

---

💛 - Coveralls