fix: update backport for changes in Numba, NumExpr, Arrow, and RDataFrame

[ianna]

address issue #2734

[jpivarski]

With or without this PR (that is, main-v1 or ianna/numba-error-backport-to-v1), I get 66 test failures. They might not be related to Numba. I'm going to work through the test failures to see what can be done.

[jpivarski]

I'm self-reviewing this because I don't think there's anything a second pair of eyes can add.

• @ianna updated all Numba errors to NumbaTypeError and NumbaValueError, like fix: using numba.core.errors.Numba<Error> instead of Error in a Numba typing context. #2659.
• I updated our NumExpr interface like fix: adjust for numexpr 2.8.5, which hid getContext's frame_depth argument #2617.
• I updated our pyarrow-Parquet interface to allow .list.element in addition to .list.item, similar to fix: update string to search for in determing parquet column list separator #2670.
• I dropped failing RDataFrame tests without investigating why they were failing. This is only in the _v2 subdirectory, which no one should be using now: the 1.x backport is for using 1.x. If anyone is using 2.x interfaces, they should be using Awkward 2.x.

After merging this into main-v1, I'll also publish a release as 1.10.5.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (main-v1@e68a046). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head 084f90e differs from pull request most recent head 4492cdc. Consider uploading reports for the commit 4492cdc to get more accurate results

Additional details and impacted files

[jpivarski]

I know that the Windows 32-bit wheels build and are successfully tested from those builds.

https://github.com/scikit-hep/awkward/actions/runs/6410147455/job/17402806110?pr=2735

There is no reason to struggle with build-test.yml, to get its Windows to see the correct version of Python when wheels.yml does it correctly.

So I'll just remove the 32-bit test from build-test.yml and merge.

[ianna]

I like the −68,438 :-)

[jpivarski]

Oh yeah, I ran into a few problems with v2 tests failing, and this is v1, so I just dropped all of the tests. In principle, I could drop the awkward._v2 source code as well, but I didn't do that because I wasn't sure whether it would break anything. (I'd have to look more closely.)

You had me wondering, for a moment, whether there was an integer wrap-around error, since -68,438 looks a bit like -2**16.