Fix Evolution.generator

[AlbertMitjans]

• Generalize np.real to qml.math.real to make sure it works with all interfaces.
• Cast coefficients to real to avoid a warning.
• Readd features removed in Undeprecate ApproxTimeEvolution #3797

[github-actions]

Hello. You may have forgotten to update the changelog!
Please edit doc/releases/changelog-dev.md with:

• A one-to-two sentence description of the change. You may include a small working example for new features.
• A link back to this PR.
• Your name (or GitHub username) in the contributors section.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@ed4b440). Click here to learn what that means.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##             master    #3796   +/-   ##
=========================================
  Coverage          ?   99.73%           
=========================================
  Files             ?      344           
  Lines             ?    30154           
  Branches          ?        0           
=========================================
  Hits              ?    30075           
  Misses            ?       79           
  Partials          ?        0           

Impacted Files	Coverage Δ
pennylane/ops/op_math/evolution.py	97.43% <100.00%> (ø)
pennylane/ops/op_math/exp.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[rmoyard]

It looks great @AlbertMitjans 👍 Just one question and also it would be great to add a test for interfaces and coefficient

[rmoyard]

    @property
    def grad_method(self):
        return self.base.grad_method if hasattr(self.base, "grad_method") else super().grad_method

Is it needed in this PR @albi3ro ?

[rmoyard]

Thanks @AlbertMitjans 💯

[timmysilv]

I'm not sure what prompted this PR, can someone give an example of what causes this warning? if it's because the user provided something that might result in something that requires truncating some produced real part of a coefficient, idk if hiding a warning is the best move. If it's just because of weird typing things and a result of pennylane implementation details, this lgtm 👍

[rmoyard]

@timmysilv We were using plain numpy to do operations (real) on array boxes which are autograd objects. We need to make that more general and use the math module.

[timmysilv]

still not 100% sure what warning we're avoiding. i'm guessing it's a teeny tiny real part sneaking into the coeff and we're just filtering it out, which is good for me. either way it looks good