Document how java.util.concurrent.ForkJoinPool.common.* system properties should be set for Druid nodes

[leventov]

See the class-level Javadoc for ForkJoinPool:

• java.util.concurrent.ForkJoinPool.common.parallelism: I think it should be N_CPU / 2 on historical and broker nodes and realtime peons (note that the default is N_CPU) to leave more CPU resources to query-processing threads, and N_CPU - 1 (or - 2?) on other types of Druid processes (Coordinator, MiddleManager, Overlord, Router).
• java.util.concurrent.ForkJoinPool.common.exceptionHandler - probably we should create a Druid-specific exception handler that does the right thing in terms of logging, alerting, etc (or such ExceptionHandler already exists in the codebase?) but we can't configure it programmatically from inside a Druid process, it must be configured via the system property before the Druid process has started.

Related to #7143.

FYI @gianm @jihoonson @drcrallen

[drcrallen]

IMHO this shouldn't be a fork join pool specific thing, but part of a larger "here are your CPU thread pools" kind of documentation that includes jetty pools, http client pools, etc.

Processing and ForkJoinPool.common sounds like a great start for such a doc and worth getting in even it if is "just" those two and passive mention to others.