[BACKPORT-0.6][Bugfix][Arith] keep div_mode during floordiv simplify

[yzhliu]

@tqchen @sergei-grechanik please take a look.

[tqchen]

Thanks @yzhliu , seems we need a different testcase, that uses a truncdiv but still triggers DivModeCompatibleTo

[ANSHUMAN87]

@yzhliu : The original issue is fixed indeed, Thanks!
I think it would be good if we add the issue test case as well.

--> expr = (((x0 - floordiv((0 - (x15)), 2)) - 1) + floordiv((37 + ((x1 + 7)-5)), 2))

[ANSHUMAN87]

@tqchen, @yzhliu : I have one concern here(However not related to this PR).
As you see the expr below:
expr = (((x0 - floordiv((0 - (x15)), 2)) - 1) + floordiv((37 + ((x1 + 7)-5)), 2))

should be evaluated to actually
expr = x0

but with current simplify stage(rewrite_simplify + canonical_simplify), it evaluates to below:
expr = ((floordiv((0 - (x1: int325)), 2) + x0: int32) - floordiv((x1-5), 2))

But when i run stages as (rewrite_simplify + canonical_simplify + rewrite_simplify) it evaluates perfectly to x0.

So may be we need to run these stages with some customized config every time or until there are no more changes?

[yzhliu]

@ANSHUMAN87 good catch, this is what the argument step for I added in #5618

[tqchen]

@yzhliu please also send a patch to v0.6