RUBY-2132 improve connection reconnection after fork

[p-mongo]

https://jira.mongodb.com/browse/RUBY-2132

• Remove pid checks from non-monitoring connections and place them into connection pool. This permits the pool to close/discard connections from the parent process in the child, instead of these connections getting reconnected in place.
• When operation on connections fail with network errors, mark connections as errored and do not check them into pools.
• The monitoring connections still need the pid check for the time being, moved it into the monitoring connection from connection base.
• Add an integration test for using monitoring connection after fork (this is transparently supported for now).
• Add an integration test for using a client after fork without reconnection. This does work but new non-monitoring connections are created for operations by the pool.
• Add a stress test where multiple forked children are performing operations on the client created in the parent, without reconnecting. This intermittently fails in evergreen due to I assume network connectivity issues in ec2. Set this test to retry twice.
• Add a stress test where the parent performs operations while forked children are also performing operations on the same client without reconnecting. Operations need to be retryable by the driver (e.g. reads) to not have any application errors.
• All of the fork tests are skipped on jruby.
• Connection generation number added to command started events for diagnostics, and to error notes for the same reason.
• Unit tests that mock two consecutive ismaster failures may need to be revisited in the future, since this no longer works naively due to ismasters being sent on different driver connections.