segault logging could be more robust

[travisdowns]

If a crash of the type described in #2697 occurs, we don't log the backtrace, because ::backtrace crashes in the handler (this the same kind of crash that triggered the original segfault in this case).

Some of this was addressed in #2691 but we can we can make the segfault logger more robust by logging the segfault frames incrementally, which will help pinpoint the crash.