[BUG]: host-only get_global_resource() may be called from GPU-side code.

Is this a duplicate?

I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Compile-time Error

Component

Thrust

Describe the bug

During investigation of UB issues in thrust we've ran into a peculiear case of ptxas failing due to an unresolved reference:
#2813 (comment)
#2813 (comment)

The bottom line is that a host-only function get_global_resounce here:

cccl/thrust/thrust/mr/memory_resource.h

Line 207 in f4d358a

_CCCL_HOST MR* get_global_resource()

Gets called from a host/device constructor here:

cccl/thrust/thrust/mr/allocator.h

Lines 203 to 206 in f4d358a

    
           _CCCL_EXEC_CHECK_DISABLE 
        
           _CCCL_HOST_DEVICE stateless_resource_allocator() 
        
               : base(get_global_resource<Upstream>()) 
        
           {}

In some scenarios, we end up trying to call it from the GPU-side code. To add insult to injury neither NVCC nor clang currently diagnose such an invalid call. Clang bug: llvm/llvm-project#118415

How to Reproduce

The issue was originally seen during compilation of cub/test/catch2_test_device_three_way_partition.cu
#2813 (comment)

Expected behavior

If the function may be called from a GPU code, it should have appropriate attributes.

Reproduction link

No response

Operating System

No response

nvidia-smi output

No response

NVCC version

No response

	_CCCL_EXEC_CHECK_DISABLE
	_CCCL_HOST_DEVICE stateless_resource_allocator()
	: base(get_global_resource<Upstream>())
	{}