Reduce contiguous-split block_size for copy_partition (#8216)

The `cudf::contiguous_split` API fails with error `cudaErrorLaunchOutOfResources too many resources requested for launch` when launching the internel `copy_partition` kernel. 
https://github.com/rapidsai/cudf/blob/c2c67de0a487ef767f2b16e6b95132121e2eec04/cpp/src/copying/contiguous_split.cu#L1024-L1026

Was determined that the `block_size=512` was too large for the debug build to launch the kernel. Changing this to `256` solved the error and all split_tests are passing. Ran the gbenchmark CONTIGUOUS_SPLIT_BENCH (on a Release build) and found no measurable performance difference running on my Linux desktop with a GV100. I can post the benchmark results if desired.

So this PR just changes the hardcoded block size for the `copy_partition` kernel from 512 to 256.

Authors:
  - David Wendt (https://github.com/davidwendt)

Approvers:
  - Mark Harris (https://github.com/harrism)
  - Jason Lowe (https://github.com/jlowe)
  - https://github.com/nvdbaranec

URL: #8216

cpp/src/copying/contiguous_split.cu

@@ -1021,7 +1021,7 @@ std::vector<packed_table> contiguous_split(cudf::table_view const& input,
 
   // copy.  1 block per buffer
   {
-    constexpr size_type block_size = 512;
+    constexpr size_type block_size = 256;
     copy_partition<block_size><<<num_bufs, block_size, 0, stream.value()>>>(
       num_src_bufs, num_partitions, d_src_bufs, d_dst_bufs, d_dst_buf_info);
   }