Port all conda recipes to rattler-build

[gforsyth]

Port all condabuild recipes over to use rattler-build instead.

Contributes to rapidsai/build-planning#47

• To satisfy rattler-build, this changes all the licenses in the pyproject.toml files to the SPDX-compliant Apache-2.0 instead of Apache 2.0

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[gforsyth]

/ok to test

[gforsyth]

/ok to test

[bdice]

Tacking on my changes from #656 into this PR. Please take these suggestions with a grain of salt -- I haven't fully validated them, and there may be some gray area here where we need to support libcufile on ARM and are willing to accept some minor issues on CUDA 12.0/12.1 in order to get the benefits of libcufile on CUDA 12.2+.

[gforsyth]

This is the same failure I was seeing before:

=========================== short test summary info ============================
FAILED tests/test_cufile_driver.py::test_open_and_close - RuntimeError: KvikIO failure at: /tmp/conda-bld-output/bld/rattler-build_libkvikio/work/cpp/src/shim/cufile.cpp:99: KvikIO not compiled with cuFile.h
FAILED tests/test_cufile_driver.py::test_property_accessor - RuntimeError: KvikIO failure at: /tmp/conda-bld-output/bld/rattler-build_libkvikio/work/cpp/src/cufile/driver.cpp:181: KvikIO not compiled with cuFile.h
= 2 failed, 1350 passed, 105 skipped, 10 xfailed, 2 xpassed in 71.12s (0:01:11) =

On the arm job testing against cuda 12.0.1

So I think I'll revert the linux64 changes to keep parity with the existing recipe for now, although I would love to not deal with the nvidia channel

[gforsyth]

Hey @bdice -- I think I've got a reasonable compromise in-place here if you want to take a look

[bdice]

Approving the specific concerns I had about libcufile on ARM. Sticking to the status quo for now and revisiting later seems like the best we can do.

[gforsyth]

/merge