Migrate ML backends to Qiboml

[MatteoRobbiati]

Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.82%. Comparing base (ad50a20) to head (750a510).

❗ Current head 750a510 differs from pull request most recent head 72eea4b

Please upload reports for the commit 72eea4b to get more accurate results.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1365      +/-   ##
==========================================
- Coverage   99.84%   99.82%   -0.02%     
==========================================
  Files          74       72       -2     
  Lines       10806    10507     -299     
==========================================
- Hits        10789    10489     -300     
- Misses         17       18       +1     

Flag	Coverage Δ
unittests	99.82% <100.00%> (-0.02%)	⬇️

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

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

[MatteoRobbiati]

@BrunoLiegiBastonLiegi @Edoardo-Pedicillo now tests are passing.

[BrunoLiegiBastonLiegi]

Thanks, this is fine, my only concern is whether it is better to specialize the backend loading for qiboml as you are doing here, which allows not to brake all the previous code that was using set_backend("tensorflow"), but creates an asymmetry with the other backend providers, such as qibojit. Or, whether it is preferable to modify the qiboml metabackend to have a platform argument selecting the actual backend out of the ones qiboml provides, this would simplify this PR and maintain the symmetry with the other repos, but will involve setting the backends as set_backend("qiboml", platform="tensorflow"). What do you think @alecandido ?

[alecandido]

@BrunoLiegiBastonLiegi I'd just go for the uniform scheme, no need to make Qiboml different from the other providers.

We should keep a single instruction, just for "tensorflow", that will check explicitly for it, and, in case, replace the backend with "qiboml", and the platform with "tensorflow".
This should be just a transition device, until we decide to break for good (thus, we'll open an issue about, and schedule for removal).

Moreover, I'd even drop all the QIBO_NON_NATIVE_BACKENDS specification, just rely on the package name (as we're already doing).
If the package is not available, we could catch the error and re-raise, hinting that is not installed, or misspelled (that could even be non-existing).
This will allow future backend providers to avoid at all the need of editing qibo source.