Fix JSON Schema generation for union dtypes

[sneakers-the-rat]

Fix the json schema half of: #38

[coveralls]

Pull Request Test Coverage Report for Build 12325636494

Details

• 14 of 15 (93.33%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.05%) to 98.298%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/numpydantic/testing/helpers.py	13	14	92.86%

Totals
Change from base Build 12169180503:	-0.05%
Covered Lines:	1502
Relevant Lines:	1528

---

💛 - Coveralls

[sneakers-the-rat]

this is the damndest thing... I can replicate #38 when i run just the pipe union tests like

python -m pytest -m pipe_union -k "jsonschema" -v

but i can't replicate it if i select any other cases - here the union marker is applied to all the pipe_union tests (i.e. union is a superset of pipe_union):

python -m pytest -m union -k "jsonschema" -v

or generically

python -m pytest -k "jsonschema" -v

pydantic must be doing some schema caching.....

[sneakers-the-rat]

oh omg this is yet another nptyping bug.

apparently hash(typeA | typeB) == hash(Union[typeA, typeB]) despite one being a types.UnionType and the other being a typing.Union..... so since nptyping caches types instead of creating new ones (for some reason???) it masks the error in the tests

[sneakers-the-rat]

I think the reason nptyping doesn't create a new type every time is for type equality comparisons, so i'm preserving that for now by adding additional type information in the key for the dict that stashes them, but these will go away in 2.0 when we finally are rid of nptyping

[sneakers-the-rat]

ah, that's better, nothing like correctly failing tests to make you feel sane again