Add Scalar backend

[jeffry1829]

Not fully tested, at least compile passed.

It's important to review SProxy, since we're not using straightforward way to write into memory.
Instead, we first convert memory to at::Tensor, then use tn[loc] = elem;

[codecov]

Codecov Report

Merging #306 (230a6b8) into dev-master (dd0fdbb) will increase coverage by 3.61%.
Report is 38 commits behind head on dev-master.
The diff coverage is n/a.

@@              Coverage Diff               @@
##           dev-master     #306      +/-   ##
==============================================
+ Coverage       13.75%   17.37%   +3.61%     
==============================================
  Files             282      274       -8     
  Lines           55158    57728    +2570     
  Branches        24712    26747    +2035     
==============================================
+ Hits             7588    10030    +2442     
+ Misses          41574    39143    -2431     
- Partials         5996     8555    +2559     

see 70 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[kaihsin]

This need to be careful to not mess-up the gradient part.

[kaihsin]

is it possible to not using switch? I think torch must have their own function you can call?

[jeffry1829]

Since we're dealing with cytnx types. I think we cannot use pytorch.