Consider refreshing TableDirectoryListing on WatchService notification instead of 1-minute interval

[glukos]

TableDirectoryListing#refresh can be a heavy operation, especially if a frequent roll cycle is chosen. e.g. RollCycles#MINUTELY. Even though the allocation issue is mitigated under https://github.com/OpenHFT/Chronicle-Queue/issues/922, calling it once a minute may cause latency spikes. We should consider registering a WatchService in queue in order to trigger refresh only on start or if queue files were manually moved.