Patch hb_set implementation back to pre 1.6.1 implementation

Perf bisects indicate that the hb_set changes may be the reason for
seing performance regressions in issues 781794 and 782220. This is
somewhat speculative as there were issues with reproducibility in the
perf_bisect runs. Results from the main perf bot graphs should help
figure out whether the hb_set changes were the culprit here.

Bug: 782220, 781794
Change-Id: I5462b37503ca76a299c99cf41388d0d195722358
Tbr: behdad@chromium.org
Reviewed-on: https://chromium-review.googlesource.com/804374Reviewed-by: Dominik Röttsches <drott@chromium.org>
Commit-Queue: Dominik Röttsches <drott@chromium.org>
Cr-Commit-Position: refs/heads/master@{#521069}

third_party/harfbuzz-ng/README.chromium

@@ -24,3 +24,7 @@ Locally added #undef PAGE_SIZE to hb-set-private.hh to avoid a collision with
 
 Temporarily patched to original CoreText default font size 36.f that was used
 before to ease Blink reabaselining.
+
+Locally patching hb_set implementation back to the pre 1.6.1 implemention to see
+whether this addresses reports of performance regressions in issues 781794,
+782220.

third_party/harfbuzz-ng/src/hb-set.cc

@@ -45,8 +45,7 @@ hb_set_create (void)
   if (!(set = hb_object_create<hb_set_t> ()))
     return hb_set_get_empty ();
 
-  set->page_map.init ();
-  set->pages.init ();
+  set->clear ();
 
   return set;
 }
@@ -96,8 +95,7 @@ hb_set_destroy (hb_set_t *set)
 {
   if (!hb_object_destroy (set)) return;
 
-  set->page_map.finish ();
-  set->pages.finish ();
+  set->fini ();
 
   free (set);
 }
@@ -378,12 +376,11 @@ hb_set_symmetric_difference (hb_set_t       *set,
  * 
  *
  * Since: 0.9.10
- *
- * Deprecated: 1.6.1
  **/
 void
 hb_set_invert (hb_set_t *set)
 {
+  set->invert ();
 }
 
 /**