Network and command errors encountered in multi-threaded monitoring are logged as debug errors (with level MONGOC_LOG_LEVEL_DEBUG). They should be logged as MONGOC_LOG_LEVEL_ERROR and MONGOC_LOG_LEVEL_WARNING.

They were left as debug logs in CDRIVER-3535. This was because error/warning logs are printed to stderr. The test script compile-unix.sh asserts that tests nothing is printed to stderr. But many tests simulate monitoring errors with the mock server. See this conversation in the PR for more context. https://github.com/mongodb/mongo-c-driver/pull/625/files/65977607f7e1d500dfa4123711452c435ab08d3c#r435400372

We may need to add calls to capture_logs() in all tests where we expect monitoring errors to occur, so they do not get printed to stderr.

Also consider:

1. Errors encountered in single-threaded monitoring are not logged. If these are generally useful, it follows we should add similar logs for single-threaded monitoring in mongoc-async.c.
2. have test-libmongoc make assertions that unexpected error logs are not encountered. This way, it will be easier to test these assertions locally, and on any platform.