Build conda package

[multimeric]

As discussed with @pr4deepr.

Advantages:

• Faster installation compared to pip (I think)
• Can automatically resolve the OpenCL and other system dependencies

Blockers: aicspylibczi and npy2bdv apparently don't have conda packages. Looking into this.

How it will work:

1. I'll add a meta.yaml recipe to core/ and plugin/
2. In the short term, we can then build the conda package from that using `conda-build
3. The conda package can then get published to GitHub
4. Later we can submit it to bioconda or conda-forge so that it can be more easily installed

[multimeric]

Currently waiting on these submitted conda PRs:

• aicspylibczi conda package
• npy2bdv conda package
• Fix aicsimageio package
• Fix ome-types package

[multimeric]

This is more difficult than it seems, because not every PyPI package version is available on conda. I had to create packages for aicspylibczi and npy2bdv from scratch, but other packages are missing crucial versions that we need. I will continue looking, although it's not clear how viable this will be.

[multimeric]

Okay, everything is fixed upstream now, so it will be possible to build a conda package. I've messed around with building this in our own CI, and it builds successfully. However in the long run I think it makes sense to submit this as a regular bioconda or conda-forge recipe, because installing conda packages from files is tricky and annoying. So assuming this is all good, I will submit the recipe once we have published the pip package.

[pr4deepr]

Thanks @multimeric. Looks great!
How do we test this locally?
Any advantages to bioconda or conda-forge?

[multimeric]

You basically can't test it locally, which is annoying. I think we would just submit it to the repository and then test out the installation after. It won't get published unless the package works though.

I have no real preference. conda-forge is enabled by default in almost every installation, while bioconda isn't. bioconda might give slightly better discoverability for those in computational biology, except that it seems like most of the bioimaging ecosystem uses conda-forge.

[pr4deepr]

Happy with conda-forge..

[multimeric]

I created a conda package metadata file here: https://github.com/BioimageAnalysisCoreWEHI/napari_lattice/blob/f4f3bbf9facb10fa829550a34cbf9cd05e9fb96f/core/meta.yaml. This can be used to submit to conda-forge, although I don't have time to do it myself.

The one outstanding question is how to handle the ocl stuff in conda, since the dependencies will vary based on the GPU or CPU hardware you have.