Extend index read stats to include disk time and #from page cache

It would be useful to time the disk load times separately from any node decompression.  Also count the number of loads that take longer than a threshold (default 50us).

• It is useful to know how much time is spent in loading v decompression to diagnose if the process is disk or cpu bound
• The number of loads that take less than a threshold can be assumed to be returned from the linux page-cache, so it gives a measure of how effective that is.  The number that are actually loaded is relevant for the effectiveness of an external cache (e.g. hpc-cache.)