bound regions and resolving inferable types

[nikomatsakis]

I think there is an error lurking in the universally_quantify_*() routines: they do a shallow structure_of but maybe they need to do something deeper? Actually it is probably ok, but think it over in any case.

[pcwalton]

Closing this, please reopen if necessary