Make @device_code_sass work with non-Julia kernels.

[maleadt]

By using CUPTI and profiled module callbacks, instead of hooking into GPUCompiler, we can also reflect on SASS code from external libraries:

julia> @device_code_sass CUBLAS.scal!(1, 1f0, CUDA.ones(1))
	.headerflags	@"EF_CUDA_TEXMODE_UNIFIED EF_CUDA_64BIT_ADDRESS EF_CUDA_SM86 EF_CUDA_VIRTUAL_SM(EF_CUDA_SM86)"
	.elftype	@"ET_EXEC"


//--------------------- .text._Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E --------------------------
	.section	.text._Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E,"ax",@progbits
	.sectioninfo	@"SHI_REGISTERS=18"
	.align	128
        .global         _Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E
        .type           _Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E,@function
        .size           _Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E,(.L_x_92 - _Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E)
        .other          _Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E,@"STO_CUDA_ENTRY STV_DEFAULT"
_Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E:

.text._Z15scal_kernel_refI7double2dEv19cublasScalParamsRefIT_T0_E:
...

[codecov]

Codecov Report

Attention: 3 lines in your changes are missing coverage. Please review.

Comparison is base (d8c6d6d) 71.32% compared to head (3e11958) 71.33%.

Files	Patch %	Lines
src/compiler/reflection.jl	86.36%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2247      +/-   ##
==========================================
+ Coverage   71.32%   71.33%   +0.01%     
==========================================
  Files         155      155              
  Lines       14837    14854      +17     
==========================================
+ Hits        10582    10596      +14     
- Misses       4255     4258       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.