Use sccache-dist build cluster for conda and wheel builds

[trxcllnt]

Description

RAPIDS has deployed an autoscaling cloud build cluster that can be used to accelerate building large RAPIDS projects.

This PR updates the conda and wheel builds to use the build cluster.

This contributes to rapidsai/build-planning#228.

[pentschev]

Why are we adding NVCC flags here? UCXX shouldn't need NVCC to compile.

[trxcllnt]

I made the conda recipe envvar updates consistent across all the PRs, mostly because I wasn't sure which recipes use which compilers.

When using the build cluster, the rapids-configure-sccache script from gha-tools sets PARALLEL_LEVEL=<a_very_large_number> and NVCC_APPEND_FLAGS=-t=100 to maximize total parallelism.

Since it's just an envvar, it doesn't hurt to include it here. If UCXX ever does add any nvcc targets, you won't have to worry about needing to update the recipe in this way.

[trxcllnt]

Does anyone know what's going on here? I wasn't seeing this test fail a few days ago, but now it happens consistently and AFAIK I didn't change anything that should cause this. Maybe a UCX update broke something?

[trxcllnt]

I think the issue is that cudf.datasets.timeseries() is producing bad data:

(base) root@831ed948ef31:/# pip install --extra-index-url https://pypi.anaconda.org/rapidsai-wheels-nightly/simple 'cudf-cu13==25.12.*,>=0.0.0a0'
(base) root@831ed948ef31:/# python
Python 3.13.9 | packaged by conda-forge | (main, Oct 22 2025, 23:33:35) [GCC 14.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import cudf
>>> cudf.datasets.timeseries()
                       id  name     x     y
timestamp                                  
2000-01-01 00:00:00  <NA>  <NA>  <NA>  <NA>
2000-01-01 00:00:01  <NA>  <NA>  <NA>  <NA>
2000-01-01 00:00:02  <NA>  <NA>  <NA>  <NA>
2000-01-01 00:00:03  <NA>  <NA>  <NA>  <NA>
2000-01-01 00:00:04  <NA>  <NA>  <NA>  <NA>
...                   ...   ...   ...   ...
2000-01-30 23:59:56  <NA>  <NA>  <NA>  <NA>
2000-01-30 23:59:57  <NA>  <NA>  <NA>  <NA>
2000-01-30 23:59:58  <NA>  <NA>  <NA>  <NA>
2000-01-30 23:59:59  <NA>  <NA>  <NA>  <NA>
2000-01-31 00:00:00  <NA>  <NA>  <NA>  <NA>
[2592001 rows x 4 columns]

[trxcllnt]

Should be fixed by rapidsai/cudf@7d54b71

[trxcllnt]

Merged after discussing with @vyasr, who said cudf is unlikely to have a fix today.

For context, this is the exact backtrace of the failing tests:

Traceback (most recent call last):
  File "/test.py", line 106, in <module>
    asyncio.run(test_send_recv_cudf(lambda cudf: cudf.datasets.timeseries()))
    ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.13/asyncio/runners.py", line 195, in run
    return runner.run(main)
           ~~~~~~~~~~^^^^^^
  File "/opt/conda/lib/python3.13/asyncio/runners.py", line 118, in run
    return self._loop.run_until_complete(task)
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/opt/conda/lib/python3.13/asyncio/base_events.py", line 725, in run_until_complete
    return future.result()
           ~~~~~~~~~~~~~^^
  File "/test.py", line 97, in test_send_recv_cudf
    assert_eq(res, msg)
    ~~~~~~~~~^^^^^^^^^^
  File "/opt/conda/lib/python3.13/site-packages/cudf/testing/testing.py", line 860, in assert_eq
    tm.assert_frame_equal(left, right, **kwargs)
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.13/site-packages/pandas/_testing/asserters.py", line 1303, in assert_frame_equal
    assert_series_equal(
    ~~~~~~~~~~~~~~~~~~~^
        lcol,
        ^^^^^
    ...<12 lines>...
        check_flags=False,
        ^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/opt/conda/lib/python3.13/site-packages/pandas/_testing/asserters.py", line 986, in assert_series_equal
    assert lidx.freq == ridx.freq, (lidx.freq, ridx.freq)
           ^^^^^^^^^^^^^^^^^^^^^^
AssertionError: (None, <Second>)

[trxcllnt]

rapidsai/cudf#20709 should fix the failing UCXX test