From the original PR:
Loading 1000 static listeners in a binary built with gcc takes ~51s,
but only ~1.5s (34x faster) in a binary built with clang.
The problem was mitigated by #4117, which reduced number of preallocated elements in the LcTrie, though the difference in performance still exists (due to unpacked LcNode).
cc @mattklein123 @htuch @alyssawilk @lizan