[FEA] Allow <WindowExec> to partition on Decimal 128 when running on the GPU

[NVnavkumar]

Is your feature request related to a problem? Please describe.
Currently, the Spark Rapids plugin limits the types supported for WindowExec calls to common types and Decimal 64. I wish the plugin will support Decimal 128 in the partition spec for Window operations. This relates to the issue #4626 which was closed when StructType support was enabled in #4673.

Describe the solution you'd like
GPU support should be enabled and corresponding test should be added to test for proper support of this feature.

Additional context
libcudf grouped_rolling_window() was tested for __decimal128 support earlier, so this should work in the plugin now.