build tensorflow on the cirun GPU machine

[njzjz]

Guidelines for marking packages as broken:

• We prefer to patch the repo data (see here)
  instead of marking packages as broken. This alternative workflow makes environments more reproducible.
• Packages with requirements/metadata that are too strict but otherwise work are
  not technically broken and should not be marked as such.
• Packages with missing metadata can be marked as broken on a temporary basis
  but should be patched in the repo data and be marked unbroken later.
• In some cases where the number of users of a package is small or it is used by
  the maintainers only, we can allow packages to be marked broken more liberally.
• We (conda-forge/core) try to make a decision on these requests within 24 hours.

What will happen when a package is marked broken?

• Our bots will add the broken label to the package. The main label will remain on the package and this is normal.
• Our bots will rebuild our repodata patches to remove this package from the repodata.
• In a few hours after the anaconda.org CDN picks up the new patches, you will no longer be able to install the package from the main channel.

Checklist:

• I want to mark a package as broken (or not broken):

  • Added a description of the problem with the package in the PR description.
  • Pinged the team for the package for their input.

• I want to archive a feedstock:

  • Pinged the team for that feedstock for their input.
  • Make sure you have opened an issue on the feedstock explaining why it was archived.
  • Linked that issue in this PR description.
  • Added links to any other relevant issues/PRs in the PR description.

• I want to request (or revoke) access to an opt-in CI resource:

  • Pinged the relevant feedstock team(s)
  • Added a small description explaining why access is needed

• I want to copy an artifact following CFEP-3:

  • Pinged the relevant feedstock team(s)
  • Added a reference to the original PR
  • Posted a link to the conda artifacts
  • Posted a link to the build logs

• I want to add a package output to a feedstock:

  • Pinged the relevant feedstock team(s)
  • Added a small description of why the output is being added.

See conda-forge/tensorflow-feedstock#408 (comment). TensorFlow starts to link libcuda.so.1 which can be accessible on a GPU machine and inside its docker. This is a runtime requirement. I've validated locally and confirmed this can pass the GPU builds. I think this is the easiest way.

I am not sure whether the memory on cirun-openstack-gpu-xlarge is enough so I also request cirun-openstack-gpu-2xlarge.

cc @conda-orge/tensorflow

[jaimergp]

Replied in the linked PR.