[NDTensorscuTENSORExt] [BUG] cuTENSOR extension and block sparse contraction

[mtfishman]

@kmp5VT reported that block sparse contraction fails when the cuTENSOR extension is enabled (while it works if just the CUDA extension is enabled), likely due to some issue passing views of CuArrays to cuTENSOR. This is a reminder to investigate that issue and maybe make an issue report to CUDA.jl about it.

[kmp5VT]

I have reported the issue in cuTENSOR to CUDA.jl JuliaGPU/CUDA.jl#2407

[mtfishman]

#1485 implemented a temporary fix which runs but has some extraneous copying which we will be able to remove once JuliaGPU/CUDA.jl#2407 is fixed.