[pandas 2.0] Scaffolding for initializing pandas arrays from NumPy arrays

[wesm]

This relates to our discussions in the design documents.

• Non-strided NumPy arrays (without missing data, some issues to address here) are initialized with zero copy.
• Strided arrays are copied

Some more tests to write and implementation to do yet. In my opinion, this code shouldn't be too smart -- if you need to examine the incoming data to determine which values are null, that logic should be elsewhere. This is responsible for constructing the array containers with the right metadata.

Code review on Gerrit: https://review.gerrithub.io/#/c/298548/

[wesm]

Merging this and will keep prototyping in a new patch to avoid rebase conflicts with other branches