C++ refactoring: unique and is_unique

[ianna]

No description provided.

[jpivarski]

In PR #1117 (which will auto-merge soon), all v2 tests are being moved into a new tests/v2 directory, and the "v2" qualifier is being removed from their filenames. Please do that for the tests in this PR, too!

The other directory restructuring that I'll be doing today will be to copy high-level code into src/awkward/v2. At first, it will be commented out, and then I'll slowly uncomment things as they're made to work with v2 only. But since that involves new files, it won't interfere with any of your ongoing work.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (main@b9a40c8). Click here to learn what that means.
The diff coverage is n/a.

[ianna]

@jpivarski - there are a couple of places where for loops should be replaced by newly written kernel functions - they are marked with FIXME. Apart from that I think, it's done.

[jpivarski]

Great!

I know this is a lot of work, and I see your FIXMEs of the last issues to fix before it's done. I added to that some choices of exception class: C++ without access to pybind forced us to put a lot of different types of exceptions into only one class, ValueError (std::invalid_argument in C++), but some of the ones labeled below should instead be np.AxisError. We also want to maintain the distinction between user errors (ValueError, TypeError, np.AxisError) and internal programming errors, which are AssertionErrors (RuntimeErrors in Awkward 1, as this is what std::runtime_error produced, and that was our closest C++ equivalent).

One thing that I didn't mention below is to add type-tracer tests after each value test, which I also mentioned to @stormiestsin in #1135 (comment). They can usually be added in a semi-automated way, and they ensure that the code is Dask-ready.

Other than that, thanks for all of this; I can see the intricacy of a lot of these cases!