[FEA] Try using `__grid_constant__` in libcudf.

**Is your feature request related to a problem? Please describe.**

Some features in libcudf like the AST and nested comparators have pretty hefty kernels. The `__grid_constant__` annotation (available in CUDA 11.7 and higher) may help reduce local memory usage in these features. We build with CUDA >=11.8 so this would be supported in our builds.

https://docs.nvidia.com/cuda/cuda-c-programming-guide/#grid-constant

In short:
> If the address of a _global_ function parameter is taken, the compiler will ordinarily make a copy of the kernel parameter in thread local memory and use the address of the copy, to partially support C++ semantics, which allow each thread to modify its own local copy of function parameters. Annotating a _global_ function parameter with __grid_constant__ ensures that the compiler will not create a copy of the kernel parameter in thread local memory, but will instead use the generic address of the parameter itself. Avoiding the local copy may result in improved performance.

@jrhemstad thinks a fair amount of the register/stack usage in AST and comparator code comes from thread local copies of those structs. @bdice agrees this is worth investigating.

(Some issue text contributed by @jrhemstad.)

**Describe the solution you'd like**
Try adding `__grid_constant__` to AST kernels and nested comparators to see how the kernels' usage of local memory, registers, etc. changes.