Correct BitArray benchmark to use actual multi-bit flag arrays

[stvoutsin]

This PR changes the votable test to use proper 2D arrays for BitArray benchmark.

It replaces scalar boolean columns with actual multi-dimensional flag arrays to accurately benchmark BitArray parsing performance. The previous implementation was forcing scalar bools into BitArrays

This could be related to the issues where the benchmarks on astropy PRs are timing out.
However I'm not sure why that would be the case, as if run locally (the version before these changes) it completes within a few seconds for each test, whereas the timeouts in the workflow runs happen after 60 seconds.

In any case I've run this version locally and see these run times:

[ 0.01%] ··· votable.TimeVOTableBitArrayOptimization.time_bitarray_16bit_binary                                                                                                                            2.56±0s
[ 0.01%] ··· votable.TimeVOTableBitArrayOptimization.time_bitarray_16bit_binary2                                                                                                                           3.25±0s
[ 0.01%] ··· votable.TimeVOTableBitArrayOptimization.time_bitarray_8bit_binary                                                                                                                             2.42±0s
[ 0.01%] ··· votable.TimeVOTableBitArrayOptimization.time_bitarray_8bit_binary2                                                                                                                            3.19±0s

If we think that it is a timeout due to the row size I could reduce it down by some fraction, but perhaps better not to include it in this PR?

[pllim]

Thanks!

Related:

• Add benchmarks for VOTable binary parsing #141
• Add converters benchmark and add Bitarray column test for votable #142