EITS: bad estimate for very skewed distributions

Description

After fix for MDEV-5926, we see a regression for the filtered% column for testcase from MDEV-4350:

filtered% used to be 50%, now it's 99.22%. Both estimates are very wrong:

But the new one is even worse than before.

Environment

None

Assignee

Sergei Petrunia

Reporter

Sergei Petrunia

Labels

Fix versions

Affects versions

Priority

Major
Configure