gdb: fix slowdown during skeletonless type units processing

gdb: fix slowdown during skeletonless type units processing

My commit 6474c699a525 ("gdb/dwarf: sort dwarf2_per_bfd::all_units by
(section, offset)") introduced a pretty bad performance regression in
the "skeletonless type units" step.  I have a pretty big executable
(Blender) compiled with -gsplit-dwarf (to generate .dwo files) and
-fdebug-types-section (to generate type units).  Before the offending
commit:

    Time for "DWARF skeletonless type units": wall 29.126, user 28.507, sys 0.497, user+sys 29.004, 99.6 % CPU

... and after:

    Time for "DWARF skeletonless type units": wall 120.768, user 119.543, sys 0.651, user+sys 120.194, 99.5 % CPU

The reason for the slowdown is that add_type_unit now inserts type units
at the right place in the all_units vector to keep it sorted.  These
repeated insertions in the middle of the vector require shifting a lot
of elements and end up taking a lot of time.

This patch fixes it by doing just one sort at the end of
process_skeletonless_type_units.  The responsibility of keeping the
all_units sorted is delegated to the callers of add_type_unit.  The
other two callers call finalize_all_units right after calling
add_type_unit.

One drawback that is probably not a real one: in
process_skeletonless_type_unit, we call process_type_unit.  If something
in there needs to look up another type unit by (section, offset), it
wouldn't find it.  I don't think that's a real issue though, as type
units are typically self contained.  If a type unit needs to refer to a
type defined in another type unit, it would do so by signature, with
DW_FORM_ref_sig8.  And during the indexing phase, I don't think we even
look at the DW_AT_type of things anyway.

With this patch applied, I am back to:

    Time for "DWARF skeletonless type units": wall 29.277, user 28.632, sys 0.521, user+sys 29.153, 99.6 % CPU

I would like to cherry pick this patch to GDB 17, to avoid shipping GDB
17 with the performance regression.

Change-Id: I2a5b89ebca9e1a4e6248032e144520c9a579f47a
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=33526
Approved-By: Tom Tromey <tom@tromey.com>
Reviewed-By: Andrew Burgess <aburgess@redhat.com>