BUG: fix raising a unyt array to an array power in sensible cases

[neutrinoceros]

Trying to fix the few cases where it does make sense to raise a unyt array to a power array.
Close #522

[xshaokun]

I’m sorry for the delayed response. When

a = unyt_array([1, 2, 3])
p = unyt_array([1, 2, 3, 4])

a**p should not raise an error. However, I tried it and that’s not the case.

[xshaokun]

This should be valid only when a is a dimensionless array.

[chrishavlin]

a**p should not raise an error. However, I tried it and that’s not the case.

What's your expectation when the array shapes don't match? In your example, those arrays will raise a broadcast error with plain numpy, so I wouldn't expect the operation to work with unyt.

a = np.array([1, 2, 3])
p = np.array([1, 2, 3, 4]
a**p

ValueError: operands could not be broadcast together with shapes (3,) (4,) 

[xshaokun]

a**p should not raise an error. However, I tried it and that’s not the case.

What's your expectation when the array shapes don't match? In your example, those arrays will raise a broadcast error with plain numpy, so I wouldn't expect the operation to work with unyt.

a = np.array([1, 2, 3])

p = np.array([1, 2, 3, 4]

a**p

ValueError: operands could not be broadcast together with shapes (3,) (4,) 

Oh, I'm sorry for the typo! These two arrays are supposed to be the same, or any two dimensionless unyt_arrays with the same shape.

[neutrinoceros]

Phew, that was a wild ride, but I think I fixed it. I don't like how complex this is turning out but I'm running out of time. Let me know if you see ways to simplify the logic.

[xshaokun]

Thank you so much for your work, everything is working fine now. Since I don’t fully understand some of your implements, I’m unable to suggest ways to simplify the logic. Nevertheless, I really appreciate it!

[neutrinoceros]

Alright. @jzuhone this is currently the last bugfix on the 3.0.4 milestone. Can you have a look ?