Make sure result objects don't store references to corresponding solver objects

[pmenczel]

Description
Currently, result objects store a reference to the solver object that created them. I think this is an accident. The options field of a solver is an instance of _SolverOptions (see here), which gets stored in the result here. _SolverOptions has a _feedback field which is assigned an instance method of the solver and thus keeps a reference to the solver.

I noticed this issue when I called qsave on a result object and the resulting file was ~100 MB instead of the expected ~100 KB. Setting result.options._feedback = None before calling qsave reduced the file size by a factor ~2000. But also without pickling, I think that this is a memory leak where solver objects can't be garbage collected.

The fix I am submitting here makes results create a copy of the options passed to it, setting _feedback to None if it exists. I considered converting _SolverOptions objects to plain dicts in the result class, but that would discard a little bit of information. (Note however that in many places, the options passed to the results are plain dicts anyway, for example.) I also considered making _feedback a weak reference, but that would make pickling more complicated.

I am not sure what tests to add about this, if any.

[coveralls]

coverage: 84.849% (+0.004%) from 84.845%
when pulling 250692b on pmenczel:result-opt-no-feedback
into 454b615 on qutip:master.

[Ericgig]

Good catch, thank you for fixing.

Changing options in result to plain dict would also be fine.

I am fine to not have a test for this, but I think adding a check on the size of file created in test_qsave_qload could work.