The correction of influence is a bit excessive in`shouldBalance`

## Enhancement Task
If there is some move-peer operator which is from A->B, B's influence will increase. So PD will try to move peers from B to C.
But we don't want to balance around. So following code is useful.
https://github.com/tikv/pd/blob/d7435da069ab96b91b9e6a6686a394cce6f42390/server/schedulers/utils.go#L91-L97

ps: I think the comments here should be clearer

But I think changing positive value into negative value is a bit excessive. Consider the following scenario
> [Score] A,B,C = 5,3,1
> [Schedule] A->B, A->C
> [Score with influence] A,B,C = 3,4,2
> [want to Schedule] B->C
> [Score in `shouldBalance`] A,B,C = 3,2,2

So we can't move peer from B to C.
IMO, we can change positive value into zero

Start in server/schedulers/utils.go at the linked shouldBalance code around lines 91-97 and trace how influence changes the scores. Reproduce the A→B and A→C scenario from the issue, then verify that positive influence is treated as zero for balancing and clarify the nearby comments.

Written by the indexing model from the issue text.