Create cluster with single page access for record

[andrii0lomakin]

Now we always perform 2 random IO read operations to read records from cluster. It makes design of generation of unique RIDs very simple but have big disadvantage, random IO slowdowns data reads a lot. We have to create cluster design which:

1. Will ensure that we have "1 page access" during read of the most of the records.
2. Ensures that we have at most "2 page access" if read the rest of the records, of course if record fits in single page.
3. Ensures that during cluster full scan if record fits inside of single page only single IO operation is needed to read the record.
4. Ensures that during full cluster scan we read data using only sequential IO.

[andrii0lomakin]

will add in 3.0 once MT TX issue #1677 will be fixed.