Pin *arrow to use *cuda in run (#8651)

Follow up to PR ( #8637 )

Even though we pinned `*cuda` in `host`, this didn't carry over to the `run` dependencies. To fix this, tack on `*cuda` to `pyarrow` and `arrow-cpp` in `run`. This should fix installation errors where the CPU builds were getting pulled in unintentionally.

cc @ajschmidt8 @charlesbluca @pentschev

Authors:
  - https://github.com/jakirkham

Approvers:
  - Jordan Jacobelli (https://github.com/Ethyling)

URL: #8651

conda/recipes/cudf/meta.yaml

@@ -42,7 +42,7 @@ requirements:
     - cupy >7.1.0,<10.0.0a0
     - numba >=0.53.1
     - numpy
-    - {{ pin_compatible('pyarrow', max_pin='x.x.x') }}
+    - {{ pin_compatible('pyarrow', max_pin='x.x.x') }} *cuda
     - fastavro >=0.22.0
     - {{ pin_compatible('rmm', max_pin='x.x') }}
     - fsspec>=0.6.0

conda/recipes/libcudf/meta.yaml

@@ -42,6 +42,7 @@ requirements:
     - dlpack>=0.5,<0.6.0a0
   run:
     - {{ pin_compatible('cudatoolkit', max_pin='x.x') }}
+    - arrow-cpp * *cuda
     - arrow-cpp-proc * cuda
     - {{ pin_compatible('dlpack', max_pin='x.x') }}