[0/5] RFC - fix some "dwz -m" regressions

[Tom Tromey]

This series fixes most of the regressions seen when running the test
suite with "dwz -m".

I say "most" because, even after these changes, there are still a couple
of regressions:

    gdb.base/dbx.exp: whereis my_list: PASS => FAIL
    gdb.cp/namespace.exp: info func xyzq: PASS => FAIL

These are caused by the odd way that we invoke dwz -- namely by running
it on two identical copies of a program.  This yields weird results; not
exactly incorrect, but unlikely to appear in real-world uses.  I didn't
fix these because it seemed like these were past the point of
diminishing returns.

(break-interp.exp also fails on this machine, but this because prelink
doesn't understand the dwz-generated DWARF at all.)

I appreciate comments on the patches.
This is all post-branch stuff.

Tom