64k+-sections-change causes problems preserving section info in elf_object_p and elf_core_file_p

[Hans-Peter Nilsson <hans-peter dot nilsson at axis dot com>]

See most test-cases fail on cris-elf, objdump outputting section
data twice.  You'd probably see it with --enable-all-targets
too, excluding CRIS vectors.

This is because section info is supposed to be preserved over
calls to elf_object_p (and elf_core_file_p), so
bfd_check_format_matches can try different targets on the same
bfd.  However, it's no longer enough to preserve the sections
pointer and sections count; there's now also a section_tail
pointer and a hash-table.

For the case at hand, --target=cris-elf, there are two
non-generic elf targets, elf32-us-cris and elf32-cris.  The
match order in bfd_check_format_matches is to try elf32-us-cris
then elf32-cris.  The second call causes sections to appear
twice due to the change in section info in a bfd.  The failure
is similar to those linked at
<URL:http://sources.redhat.com/ml/binutils/2001-11/msg00061.html>.

I'll try to fix this by also preserving section_tail and
introducing a (shallow-)copy function bfd_hash_table_copy for
the sections hash table, calling it to copy the hash table into
a local copy and bfd_hash_table_free the one in bfd and
bfd_hash_table_copy it back on failure.

Other suggestions welcome.  (Fixes too. ;-)

brgds, H-P