On FreeBSD locking inside libc (e.g. for stdio) calls the libthr private
API (using _pthread_* functions)rather than the standard pthread_*.
This can lead to many false-positive race reports unless we also intercept
the version with the leading underscore.
Unfortunately this is a rather large diff since all pthread interceptors
had to be changed to use a new macro.
I don't have access to a NetBSD system for testing, but it seems like this
approach could also be used to handle the NetBSD __libc_pthread_* functions.