Import collapse to reduce the time taken by %in%

[etiennebacher]

Adds collapse::%!iin% calls instead of !%in%.

In addition to that, missing indices in a and b shouldn't be always computed (similarly to #101). For instance, if mode == "left" then not_matched_b is not used. I took the liberty of converting the multiple if else to a switch() call because I find that clearer than nested if else, but no problem to change if you prefer keeping the previous way.

Close #102

[etiennebacher]

Regarding the version in DESCRIPTION, it was 0.1.4 but the CRAN version is 0.1.2. @beniaminogreen did you use 0.1.4 as a development version? I changed that to 0.1.2.9000 which is a more common pattern to signal development versions, but again no problem to change if necessary.

[beniaminogreen]

Thanks for this, and nice catch with the development version. I did not know that that was the convention to show development versions. I'll let the tests go through and wait a few mins in case there are any other tweaks you would like to make, then I will merge it in!

[etiennebacher]

I did not know that that was the convention to show development versions

Well at least it's the tidyverse convention and many packages use it: https://r-pkgs.org/lifecycle.html#sec-lifecycle-version-number-tidyverse

Some people also bump the 9000 after every change in the development version, so that it's easier to differentiate 0.1.2.9015 and 0.1.2.9008 for example.

In any case, I'm done with this PR so feel free to merge when you want