Expose random seed fixing to developer (mostly for testing)

[quinn-dougherty]

• _ Is refactor
• x Is new feature
• _ Concerns documentation

Description of suggestion or shortcoming:

A whole bunch easier than #227, and 227 depends on it

• probably this https://www.npmjs.com/package/random
• the thing is our montecarlo often calls .sample method from Jstat, so it may not be incredibly obvious how to implement.

[berekuk]

probably this https://www.npmjs.com/package/random

Another option to investigate: https://github.com/davidbau/seedrandom

the thing is our montecarlo often calls .sample method from Jstat, so it may not be incredibly obvious how to implement.

Jstat uses Math.random (https://github.com/jstat/jstat/blob/e56dd7386e62f6787260cdc382b78b6848d21b62/dist/jstat.js#L113), so it should be possible to monkey-patch it with seedrandom. Or there's that jStat.setRandom function that could be used to avoid global monkey-patching.

[berekuk]

One interesting longer-term issue we'll have to deal with is threads. If Squiggle will become smart enough to execute in multiple threads (for better CPU utilization), we'll have to handle it somehow.

Consider:

f = getRandom()
g = getRandom()

Then:

project->runAll(~seed=123)

If Squiggle decides to execute f and g in different threads, the result will become non-deterministic even with a fixed top-level seed. We could deal with this by assigning different unique seeds to each thread, derived from the original seed.