ENH: option to check merge is one-to-one, many-to-one, one-to-many, or many-to-many

[nickeubank]

In my experience, there are few places in data work where problems with data are more evident than when merging datasets -- something that is both a problem (if you think you're doing a one-to-one merge and one of the keys isn't unique in one dataset, you can introduce huge problems) and an opportunity (checking that a merge works as expected is a great way to catch problems).

With that in mind, I'd like to propose adding a check_merge argument that accepts strings one_to_one, one_to_many, may_to_one, and many_to_many. If one of those strings is passed, the merge function would then check uniqueness of merge variables before running the merge, and if for example the merge key for the first dataset has duplicates in a one_to_many merge, it would throw an (informative) exception.

(Stata made a similar move to bake this functionality into its merge command around Stata 12 using 1:1, 1:m, m:1 syntax, which I'm also open to).

Though this functionality can be replicated with user tests, it gets tiring to write them every time...

[jreback]

see also #15068

[jreback]

I like the stat syntax, so maybe for readability accept both. Also call this check= I think. or maybe validate=

[nickeubank]

Sounds good. Will pull together a PR and ping for input.

[nickeubank]

Thoughts on whether it should check that keys AREN'T unique for the many tests?

i.e. if someone runs merge(left, right, on='a', validate="many_to_one") and it turns out that a is unique in left, should that throw an exception? I don't think Stata does, and my inclination is to say no, but open to input