We had a remarkable number of analyzer failures.

I'll fix the bots to analyze across the whole project
in a follow-up patch, that should prevent this
in the future.

@abarth