When unconfigured, PressureAwareCompactionThroughputController should sniff NIC speed and use appropriate settings

After upgrading a cluster to 2.3, we noticed that compactions were taking much longer. We tracked this down to the PressureAwareCompactionThroughputController, which became default as part of our upgrade. Our default configuration settings set the lower bound: 50MB/s, upper bound: 100MB/s. These settings make sense if you have a 1gbit NIC (~120 MB/s). For a 10gbit NIC, they’re quite inadequate.

Suggestion: instead of using fixed values as the default for these configurations, we should sniff the NIC speed and configure for an optimal value according to available bandwidth.