Raise log level of abnormal checks from INFO to >=WARN

[xealot]

Describe what happened:
We have a lot of hosts, doing a lot of checks multiple times per minute. Right now, the datadog agent log output (when set to INFO) is reaching to almost half of our actual webserver logs.

This seems ludicrous.

But we don't want to filter it out completely because there is useful information during check failures logging.

Describe what you expected:
The checks that fail should raise their logging level to WARN or better. That way, nominal messages can be ignored and abnormal messages can be captured.

Steps to reproduce the issue:
Use datadog a in a large environment.