Avoid runtime dependency on numpy when importing arro3.core.types

[3ok]

By building from source or installing the most recent pre-release on an empty virtual environment.

from arro3.core.types import ArrowSchemaExportable
# ModuleNotFoundError: No module named 'numpy'

Actually importing numpy may not be needed and we can get by with just a if TYPE_CHECKING condition and a forward reference.

[kylebarron]

Oops, this was unintentional! Thank you!

I was tired and forgot momentarily that types.py was actually imported by Python and not just by the type checker

[kylebarron]

Since we use from __future__ import annotations above, we don't need to put this in a string

Suggested change

	ArrayInput = Union[ArrowArrayExportable, BufferProtocolExportable, "np.ndarray"]
	ArrayInput = Union[ArrowArrayExportable, BufferProtocolExportable, np.ndarray]

[3ok]

Yes I thought the same but actually you still need it for aliases, otherwise I get this

>>> from arro3.core.types import ArrowSchemaExportable
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/redab/tmp/venv/lib/python3.10/site-packages/arro3/core/types.py", line 82, in <module>
    ArrayInput = Union[ArrowArrayExportable, BufferProtocolExportable, np.ndarray]
NameError: name 'np' is not defined

[kylebarron]

Ugh damn it 😆, well TIL, sorry for messing up your PR!

We should really have some minimal CI tests to ensure that the library can at least import without any external dependencies.

[kylebarron]

This is the second error I made in the same file 😅 #207