give `*_join()` more control on duplicated and unmatched rows

[mgirlich]

As @lionel- nicely proposed in #3967 it would be great to have more control on what the *_join() functions do on duplicated or unmatched rows. Proposed arguments:

• .unmatched = c("keep", "drop", "error")
• .duplicates = c("error", "product")

for example a 1:1 join would then simply be .unmatched = "error" and .duplicates = "error".

As the arguments should rather be added to the existing *_join() functions the "keep" and "drop" options for .unmatched don't really make sense for all joins. Rather, one would have

left/right/full_join() -> .unmatched = c("keep", "error")
inner_join() -> .unmatched = c("drop", "error")

[mgirlich]

@lionel- I didn't fully understand your thoughts about nesting and .duplicates = "keep". Maybe you'd like to add your thoughts here

[lionel-]

I think this was from a discussion with @hadley but I no longer fully understand it either 😅: #3967 (comment). Might not be as much a concern if we're adding these arguments to existing join functions.