Fix subgraph handling of non-contiguous nodelist argument

[mtreinish]

Summary

This commit fixes the subgraph() method on the coupling map class when
the node list had a non-contiguous list that didn't start at 0.
Previously, the subgraph method would inject nodes for the holes in the
node list so that all the indices specified in the nodelist argument
were present in the output subgraph. But this was neglecting the nodes
get reindexed and wasn't valid. This commit fixes this by just removing
that step and relying on retworkx's subgraph() method to do the
underlying work.

Details and comments

Fixes #6736

[ajavadia]

I just noticed there's a .reduce() method which does basically the same thing (without this bug). Is there a reason to have both?

[mtreinish]

I'm not sure there is a functional difference between them. I didn't even know there was a reduce() method. Looking at the code, reduce() will preserve node list order while subgraph() won't. The subgraph() implementation should be faster, but I don't expect either method to be a bottleneck. We can deprecate subgraph() in a follow up PR if you want, it would be better to merge the fix without deprecation so we can backport it for 0.18.1.

[nonhermitian]

I suspect reduce, or something equivalent, will come in handy in the future to run independent experiments on a single chip simultaneously.