Use nptyping..

[goodboy]

Would love to get some sane typing for all our numpy processing internals.

nptyping looks da bomb and I think we should give it a super serious shot 🏄🏼

In particular leveraging the struct-array facilities:
https://github.com/ramonhagenaars/nptyping#numpy-structured-arrays

---

TODO:

Just dumping a bunch of related links from bugs/research for now. I'll come back and clean this up..

• tractor has a PR we need to get cleaned up and get tested when used with numpy:

  • Shared memory array API and optional tight integration with numpy goodboy/tractor#338

• consider how all this will interplay with the desire for an FSP chaining/cascade DSL:

  • Designing a stream-composable, (soft) real-time ready, chain-time-type-validated FSP cascade system and DSL  #270

• numba related issues and research:

  • parallel facilities that might work well for doing cross-process QPath rendering
    • https://numba.pydata.org/numba-doc/latest/user/examples.html#multi-threading
    • https://numba.readthedocs.io/en/stable/user/parallel.html#numba-parallel
  • getitem access with str value to a struct array causes impl "no match" numba/numba#8622
  • Declaring Numpy structured array dtypes throws TypeError numba/numba#5753