RFC: unfixing

[st--]

It'd be convenient to be able to unfix a parameter, I think it should satisfy the following:

unfix(p) = p
unfix(p::Fixed) = p.value

plus recursively applying unfix on nested structures, not quite sure what exactly that should look like, perhaps mapping on NamedTuple/Tuple/Dict?
What do you think ?

[willtebbutt]

I think this would be helpful for the kinds of situations we discussed elsewhere. Specifically, when you want to optimise a subset of your parameters initially with a subset held fixed, and then optimise the remainder of your parameters once that initial optimisation has completed.

I would be interested in seeing a PR that does this -- would definitely want a worked example to shove in the readme too.