Update to jaxlib 0.3.15

[hawkinsp]

Closes #122

Checklist

• [x] Used a personal fork of the feedstock to propose changes
• [x] Reset the build number to 0 (if the version changed)
• [x] Re-rendered with the latest conda-smithy (Use the phrase code>@<space/conda-forge-admin, please rerender in a comment in this PR for automated rerendering)

Changes:

• Bump jaxlib version to 0.3.15. This appears to work when built locally for non-CUDA and CUDA linux builds.

• Remove patch to change bazel --version to bazel version. I don't think this patch is needed any more: bazel --version is something that all sufficiently modern versions of bazel support as best I can tell.

• Remove patch to move --config=cuda and ---config=nonccl around. I don't see any reason for this patch.

• Remove code to replace c++14 with c++17. Upstream JAX requires C++17 already.

• Change CUSTOM_BAZEL_OPTIONS to append to .bazelrc rather than pass --bazel_options. This is both more readable and allows us to limit the applicability of these flags only to the "build" phase. In my local builds, I saw build errors from "bazel shutdown" because it was being passed flags only applicable to "build".

• Remove scipy as a build-time dependency: it is needed only as a Python dependency at runtime.

• Remove python-flatbuffers as a runtime dep. JAX no longer depends on the Python flatbuffers library.

[conda-forge-linter]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[hawkinsp]

I see @ngam is trying this in https://github.com/conda-forge/jaxlib-feedstock/pull/122

This is as much as anything an experiment for me: I had no problem building locally with and without CUDA with these changes. I'm not sure what goes wrong in that case, maybe one of the build configurations I'm not testing?

[ngam]

I see @ngam is trying this in #122

This is as much as anything an experiment for me: I had no problem building locally with and without CUDA with these changes. I'm not sure what goes wrong in that case, maybe one of the build configurations I'm not testing?

Hi!!! Thank you for this --- if it passes locally, it will likely pass here. Let me quickly check where we diverged. I had many different iterations of that PR ...

[ngam]

FYI, on this CI it takes 4+ hours approx to finish. But if it fails like before, it will be fail in less than 30 mins.

[ngam]

Let's wait to see how far the cuda builds go, and then we will fix the cross-compilation of osx-arm64, but this looks promising! I already see the cuda builds progress much farther than before!!