[NDTensors][NDTensorsCUDAExt] Improve performance of GPU backends

[kmp5VT]

This PR is to address the issue #1193. In order to avoid scalar operations, fill tensors with rand or zeros on device.

[codecov-commenter]

Codecov Report

Attention: 2 lines in your changes are missing coverage. Please review.

Comparison is base (3cdba94) 85.37% compared to head (b7743bb) 67.34%.
Report is 2 commits behind head on main.

❗ Current head b7743bb differs from pull request most recent head 262e924. Consider uploading reports for the commit 262e924 to get more accurate results

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1194       +/-   ##
===========================================
- Coverage   85.37%   67.34%   -18.03%     
===========================================
  Files          88       88               
  Lines        8416     8388       -28     
===========================================
- Hits         7185     5649     -1536     
- Misses       1231     2739     +1508     

Files	Coverage Δ
src/ITensors.jl	100.00% <ø> (ø)
src/broadcast.jl	87.83% <100.00%> (+0.94%)	⬆️
src/itensor.jl	80.99% <100.00%> (-1.26%)	⬇️
src/tensor_operations/permutations.jl	96.77% <100.00%> (ø)
src/mps/dmrg.jl	61.94% <66.66%> (-22.16%)	⬇️
src/set_types.jl	0.00% <0.00%> (ø)

... and 34 files with indirect coverage changes

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