expose get_or_insert_key and some optimization function for dict

[rluvaton]

Which issue does this PR close?

None

Rationale for this change

I want to insert to dict without doing value lookup, this allow me to take advantage of some assumption I have about the data

What changes are included in this PR?

-

Are there any user-facing changes?

Yes, there are user facing changed and I think I documented them

TODO

• add tests

[tustvold]

I wonder if we're overloading more onto this builder than makes sense? If you're going to use unsafe to bypass the builder checks, you might as well just construct the dictionary array directly?

[rluvaton]

Sometimes I can't, I just have a function that gets the builder and need to append to it, also the builder make sure you don't insert duplicate values

[tustvold]

also the builder make sure you don't insert duplicate values

Right but you are just bypassing this with these methods?

I just have a function that gets the builder and need to append to it

And yet you know that it doesn't contain duplicate values? This feels exceedingly niche...

[rluvaton]

It's not that it doesn't contain duplicate values, it's that I don't insert duplicate values (basically i need to append some indices from other dictionary so this will allow me to do what I did in extend_dictonary but in a more fine grain way

[tustvold]

Perhaps we might take a step back here and file a ticket with a concrete use-case, explaining what it is you're trying to achieve and we can then go from there. As it stands this PR dramatically increases the API surface without clear justification