speedup ideas

[1fish2]

In addition to measure-and-tune experiments:

• MongoDB time series.
• For messages, measure the performance of dict vs. namedtuple vs. Python objects with slots.
  • (TypedDict instances are dict instances so they wouldn't change the runtime performance but they would improve mypy type checking and IDE code editing.)
• Cythonize some inner loops.
• Process-specific simulation time steps.
• Mongita is an in-process library like sqlite for MongoDB. This eliminates the interprocess communication and the installation and startup work, esp. good for unit tests. Is one CPU process enough for simulation testing and debugging? Mongita's performance metrics don't look compelling.

I updated wcEcoli and stochastic-arrow to Python 3.11 and new library releases. There was a big optimization effort in Python 3.11 but it hardly mattered for wcEcoli! I guess most of the computation is happening in Numpy, Cython code, and Numba-compiled code.

Numpy should soon be able to use the 64-bit version of Apple's Accelerate (BLAS) library for major speedup, esp. on Apple Silicon.