feat: impl *Assign ops for types in arrow-buffer

[waynexia]

Which issue does this PR close?

Closes #.

Rationale for this change

Impl std::ops::*Assign (like AddAssign) for i256, IntervalDayTime and IntervalMonthDayNano.

What changes are included in this PR?

• new trait impl

The first commit (95da017) contains the plain implementation of AddAssign and SubAssign as examples

Are there any user-facing changes?

Yes

[waynexia]

I run into this on migrating interval types in datafusion. AddAssign and SubAssign are required here

fn increment_decrement<const INC: bool, T: OneTrait + SubAssign + AddAssign>()

[tustvold]

Can we avoid using this

[waynexia]

please check 33b69f8

[waynexia]

The failing integration test seems unrelated

[alamb]

I wonder if we could use the unary_mut kernels in Arrow instead of relying on built in kernels

I spent some more time trying to document them recently here #5798

[tustvold]

I wonder if we could use the unary_mut kernels in Arrow instead of relying on built in kernels

I don't really follow the connection, these are operations on primitive scalars, not arrays.

[tustvold]

I think this is fine, given the types in question are Copy this doesn't make any difference, but is harmless enough

[waynexia]

I wonder if we could use the unary_mut kernels in Arrow instead of relying on built in kernels

TIL. Those mut-in-place APIs look good. I believe it can help improve UDF/Plan's implementations. As for increment_decrement, it's a helper over Interval ScalarValue, so binary_mut may not work here.

[waynexia]

By the way, I noticed that i256's layout is not fixed by #[repr(C)], is that intended? @tustvold

[tustvold]

Good spot, I thought I had fixed that, but guess I forgot - #5837