Use python threading instead of solara thread in model controller

[wang-boyu]

This seems to speed up model execution as a fix to what was reported in #2289 (comment) and following comments. But I don't really know why.

[rht]

I assume this solution also works on Colab?

[wang-boyu]

Oh sorry I've been testing it on my laptop; didn't try colab!

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
BoltzmannWealth	small	🔵 +1.0% [+0.3%, +1.6%]	🔵 +0.2% [-0.0%, +0.3%]
BoltzmannWealth	large	🔵 +23.2% [-0.4%, +69.5%]	🔵 -0.8% [-3.9%, +1.4%]
Schelling	small	🔵 +0.4% [+0.0%, +0.8%]	🔵 -0.4% [-0.6%, -0.2%]
Schelling	large	🔵 +0.3% [-0.2%, +0.8%]	🔵 +1.7% [-0.3%, +3.9%]
WolfSheep	small	🔵 -0.2% [-0.4%, +0.1%]	🔵 +0.3% [+0.1%, +0.5%]
WolfSheep	large	🔵 -0.0% [-0.9%, +0.8%]	🔵 +1.1% [-0.3%, +2.3%]
BoidFlockers	small	🔵 -1.5% [-1.9%, -1.1%]	🔵 +0.0% [-0.7%, +0.8%]
BoidFlockers	large	🔵 -1.1% [-1.5%, -0.7%]	🔵 -0.1% [-1.0%, +0.9%]

[wang-boyu]

Btw it's probably worth mentioning that it's good that play and pause buttons are merged into one (in PR #2299). Otherwise the solara app runs into trouble if users keep clicking on the play button.

[Corvince]

Wow, this runs indeed a lot faster. But I was worried about the implications of using threads, since they potentially keep running forever in the background. use_thread does a lot to prevent this, but this seems to be quite slow for that reason. I now saw that its also deprecated and there is use_task, which also allows running in the python event loop. Its equally fast compared to this PR, but potentially saves us some headaches in the future. See #2304

[Corvince]

Again, thanks a lot for showing whats possible. Its way faster now, would not have happened without your effort/noticing of bad performance

[wang-boyu]

Great! Thanks @Corvince