Fix shlib-versions for mips-linux

[Roland McGrath]

> I don't like throwing everything together.  It might also not be right
> in all situations since you very well can have functions with the same
> name in different libraries, one exported, the other not.

Indeed; this is the sort of "subtle peril" I had in mind.  I think it's the
case that we don't have any such issues at the moment, but it would surely
be confusing when they came up in the future.

> I also don't think that this is a problem so hard to detect that major
> reorganizations are appropriate.  You will find the problems quite
> easily since something simply will not run.

Well, the failure mode is going to be failing to override functions and so
e.g. getting the libc stub for flockfile when libpthread is in fact linked in.
That could be a bitch.  But I agree with the principle that this is detectable.

It should be straightforward to add a test that flags any suspect cases
after the shlibs are all built.  (A suspect case would be a given symbol
that is defined in multiple libraries where the sets of versions available
from each library do not all exactly match.)