[CUDA] Assumptions on `__nvvm_read_ptx_sreg_meow()` functions

When working on `cuda::launch` in the CCCL project, we often have compile time information about the dimensions of block/cluster/grid. When using `nvcc`, we are able to do:
```cpp
template
__global__ void kernel()
{
__builtin_assume(blockDim.x == block_dim_x);
if constexpr (block_dim_x == 1)
{
__builtin_assume(threadIdx.x == 0);
}
// ...
}
```
Which can replace the builtin variable reads with values and produce a better optimized code as a result.

When I try to do same thing with `clang-cuda`, I get a warning about the assumption having side effects and being ignored. I thought this is caused by the way Clang implements these builtin variables, but I get the same behaviour when using the corresponding `__nvvm_read_ptx_sreg_meow()` builtin functions instead.

Is there a way to make this work, or one can just not make assumptions on builtin functions?

Thank you!

Start by reproducing the warning with clang-cuda using __builtin_assume and the corresponding __nvvm_read_ptx_sreg_meow() builtins, as described in the issue. Trace how clang-cuda handles these builtin calls and determine whether assumptions are expected to be accepted; done means establishing a confirmed supported behavior or a clearly documented limitation.

Written by the indexing model from the issue text.