*: fix bug when unsigned histogram meets signed ranges in feedback

[alivxxx]

What problem does this PR solve?

Sometimes, the planner will generate signed scan ranges for an unsigned primary key.
So when feedback split the signed ranges according to a histogram, it may split ranges to [5(signed), +inf(unsigned)], and it is not valid when we call validRange and will be dropped, which will lead to the wrong scan ranges.

What is changed and how it works?

First, it fixes the cases that planner generate signed ranges for the unsigned primary key; Second, it adds checks when we split ranges and make sure that the type of histogram and ranges match.

Check List

Tests

• Integration test

Code changes

• Has exported function/method change

Side effects

• None

Related changes

• Need to cherry-pick to the release branch

[codecov]

Codecov Report

Merging #10415 into master will decrease coverage by 0.0035%.
The diff coverage is 62.5%.

@@               Coverage Diff                @@
##             master     #10415        +/-   ##
================================================
- Coverage   77.3312%   77.3277%   -0.0036%     
================================================
  Files           412        412                
  Lines         86467      86555        +88     
================================================
+ Hits          66866      66931        +65     
- Misses        14474      14488        +14     
- Partials       5127       5136         +9

[winoros]

lgtm

[qw4990]

LGTM

[qw4990]

/run-all-tests