Add __getstate__ and __setstate__ methods to allow pickling. Addresses #19

[VolkerH]

As discussed in #19, the dask arrays currently returned by ND2file.to_dask() cannot be pickled, which limits the use of these returned dask arrays.

The reason why the pickling fails is due to the cythonized _rdr. This PR adds a __getstate__ method that removes _rdr from the items to be pickled and a __setstate__ method that re-creates a _rdr from the path.

I can't build the python package as I don't have the C code for the Nikon SDK. So far I tested this by monkey-patching a pip-installed nd2 package with these two methods.

I am not sure whether the way I hanlde the .closed attribute is correct.

[tlambert03]

thank you @VolkerH!

I can't build the python package as I don't have the C code for the Nikon SDK. So far I tested this by monkey-patching a pip-installed nd2 package with these two methods.

sorry about this. I think I'm going to just include the SDK here. The folks at laboratory imaging have given permission to distribute, and it would simplify a lot of things...

test fails are build-related (this is the first PR that's come outside of this repo)... so I'll deal with those too.

[tlambert03]

@VolkerH could you merge main here? I tried to do it for you but keep running into issues

[VolkerH]

Hi @tlambert03 , done !

[VolkerH]

I can now build locally.
Might add a short test, consisting of the following steps:

• create dask array from nd2,
• pickle/unpickle dask array,
• run .compute() on a slice of unpickled dask array
• assert correct values.

Is there somewhere I can fetch the nd2 files you are using for tests?

[tlambert03]

thanks @VolkerH! Will merge now and add a test later. I need to revamp the testing suite to make at least a subset of it easier for contributors to use. but let me do that in another PR