WL#1290: Better cardinality estimations for MERGE

For now MERGE table calculates it's cardinality (or, rather, it's not
cardinality per se, but a "average number of rows per unique key value", which
I'd call R) as a average of R's of individual MyISAM tables. It can produce
highly inaccurate results.

Assuming there's a 6-row MyISAM table with R=2 and column values 1,1,2,2,3,3 and
another MyISAM table with R=2 and values 4,4,5,5,6,6 - then MERGE table will
have R=2, indeed.
But if the second table will contain 1,1,2,2,3,3 (with R=2, of course), MERGE
table will have R=4.

One of the ways to fix it is to calculate min/max values for each key and range
intersections between different MyISAM tables. R can be estimated assuming that
values are uniformly distributed in each range (R or MyISAM tables is only
meaningful in this assumption anyway).