[gpuCI] Auto-merge branch-0.16 to branch-0.17 [skip ci]

[gpuCI] Auto-merge branch-0.16 to branch-0.17 [skip ci]

Due to libcudf splitting into multiple shared libraries to fix debug builds, libcuspatial was left in a broken state. We don't fully understand why, but it seems like some combination of the linker pruning the `libcudf.so` dependency since it finds no symbols it uses in it BEFORE it adds `DT_NEEDED` entries from `libcudf.so` where it does find symbols.

Added some linker flags to temporarily fix the build while we continue investigating a more proper solution from the libcudf side.

Also fixed a minor gcc9 issue related to a copy elision.

Also fixed usage of a removed cudf Python API.

Planning to upgrade to cmake 3.18 in the near future and the CUDA_ARCHITECTURES change is a breaking change, so pinning to 3.17 policies to prevent the breakages.

Zero initialize device_uvector before scatter

Co-authored-by: ptaylor <paul.e.taylor@me.com>

This PR converts all usage of cudaStream_t in cuSpatial to rmm::cuda_stream_view, following on from rapidsai/cudf#6646 and rapidsai/cudf#6648

Also reorders stream parameters to occur before MR parameters in all functions.

Update `cudf::gather` callsite in `point_in_polygon_test_large` to account for the new out of bounds policy parameter.

Authors:
  - ptaylor <paul.e.taylor@me.com>

Approvers:
  - Mark Harris
  - Christopher Harris

URL: #328

Gpuciscripts clean and update