BENCH: Squeezing out the last dregs of performance

[HaoZeke]

Some considerations (to be expanded):

• Intel suite + MKL
  • Unclear if xtensor has an MKL calling process.. MKL is unlikely to improve much (not too many ops)
  • icc should still be helpful, with the right flags (document!)
• The Hoard memory allocator
  • This might just be useful, but maybe less so since the HashMap addition..
• The Mold linker
  • Really almost pointless to even add, we'll never link as fast as the Python runtime, and it scarcely matters... (maybe better developer ease)

Most (all) of these are optionals to be documented, not really to be changed wholesale.

[MSallermann]

I'd be interested in seeing what those do for our performance. Especially, Hoard and icc.

[HaoZeke]

Also maybe this? https://github.com/dengwirda/robust-predicate