feat: support Interval +/- Interval

[izveigor]

Which issue does this PR close?

Closes #4178.

Rationale for this change

What changes are included in this PR?

Are there any user-facing changes?

Yes

[tustvold]

Could you perhaps expand on the use-case for arithmetic between heterogenous interval types, we don't provide this for other types, instead leaving this to type coercion logic? You can't for example add Int16 to Int32

[izveigor]

Hello, @tustvold!
I agree with the fact, that cast function can replace add_dyn between different types in the same group (like Int16 + Int32).
But, the cast function does not support Intervals. See the problem: #4181
Even if the implementation is done in code, in my opinion, adding different types of intervals can make life easier (for example, if you need to add a day (DayTimeType) to the year (YearMonthType), you get a ready-made type with ready-made calculations, instead of translating two data types into a third (MonthDayNanoType) and producing calculations)

[tustvold]

I will try to find time to review #4182 tomorrow, I think this is the best path forward for this, as it avoids the somewhat unclear semantics of heterogenous interval addition, whilst also being less code and more consistent.

you need to add a day (DayTimeType) to the year (YearMonthType)

But should the output be DayTimeType, potentially overflowing and having to make assumptions about days in a month, or YearMonthType, potentially truncating, or should it be MonthDayNano? I suspect arguments could be made for all three, its simpler and more consistent to just not take a side 😅

[izveigor]

Your argument seems reasonable. "Cast method" gives user more options than "Add method".
Thanks for your work, @tustvold!