Lock stdout before printing an error message to stderr

[film42]

This avoids interleaving error messages messages when search is writing to stdout.

Closes #1941

There are probably a few other places we'd want to make a similar patch.

[BurntSushi]

There are probably a few other places we'd want to make a similar patch.

I believe the only other place is in the logger? Maybe it would make sense to write our own eprintln_locked macro that does this, and then just replace every call to eprintln! in crates/core with that new macro.

I think this patch is right, but it's an unfortunate abstraction violation. This only works because the buffer printer in search_parallel will acquire the same stdout lock, which is done inside of termcolor. To avoid the abstraction violation, I think we'd need another mutex in crates/core itself, but I'm not so sure it's worth it.

To mitigate the abstraction violation, could you make the comment added here mention and acknowledge the abstraction violation? Basically what I said above. And if we put this in its own macro, we only need to write this comment once.

[film42]

Adding a new macro certainly cleans things up a lot. I added a comment in the code (but not as part of the public documentation) since a comment about an abstraction violation may not be useful to the user outside of the code. Happy to make that part of the public docs though.

In my first cut, I added a lazy static rwlock and mapped the read to STDOUT with write mapping to STDERR to keep logging decently quick even with a mutex, but ultimately, locking STDOUT was a lot cleaner and intuitive. It also accomplishes the same thing.