Adjustments to pool allocator sizes

This is end result of an experiment with different (much bigger in general) sizes of blocks inside pool allocators. It turned out that it makes no difference (other then being reflected in statistics showing much less blocks allocated, released or block switches during work) - the full replay was just 3 minutes shorter than version with default 4k objects per pool block. Moreover the price was higher allocation overhead and much higher initial allocation (unit tests that run with 64MB shm started to require 256MB, since account related indexes alone allocated over 35MB right off the bat).

So the changes (pool allocator statistics added to performance reports) and increased pool sizes were mostly reverted and didn't make it to this MR. Only change that remains is small block sizes for indexes hosting objects that are of fixed number (mostly singletons).