Fix multiqubit GPU ops not called by JitCustomBackend

[mlazzarin]

It looks like we forgot to update the call to the multiqubit GPU kernels in JitCustomBackend after we moved the latter from qibo to qibojit.

[codecov]

Codecov Report

Merging #41 (25999f1) into main (4c41a1d) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main       #41   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            9         9           
  Lines          983       978    -5     
=========================================
- Hits           983       978    -5     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/qibojit/custom_operators/__init__.py	100.00% <100.00%> (ø)
src/qibojit/custom_operators/gates.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4c41a1d...25999f1. Read the comment docs.

[mlazzarin]

I also made some changes in gates.py to improve readability. Let me know if you agree.

[stavros11]

Thanks for finding this! Both the fix and the readability changes look good to me.

I just hope that the previous fallback to CPU did not have any effect to the tests/benchmarks we were doing (thinking that we run on GPU but actually use CPU), although at least for the benchmarks I believe we would notice the difference from the results, particularly for high qibut numbers. Now it works in my machine and I can confirm that the multi-qubit GPU kernel is called during testing.

[scarrazza]

@mlazzarin thanks for these updates and fixes, looks good to me.

[mlazzarin]

I just hope that the previous fallback to CPU did not have any effect to the tests/benchmarks we were doing (thinking that we run on GPU but actually use CPU), although at least for the benchmarks I believe we would notice the difference from the results, particularly for high qibut numbers.

@stavros11 I think the benchmarks of PR #23 and PR #37 are fine, because the fallback was removed in commit 58e1dfe and re-introduced only in commit 3183df2, and the benchmarks were performed in the middle of the two commits. We may need to re-run the tests with the different configurations, though, but I usually do it every time a create a new environment, so it should be fine.