I originally thought usingshutdown_script would be better, but after some testing it seems likeshutdown_scripts don't get run nearly as consistently. Idk what's going on there, but running this on startup shouldn't be too bad because it should always be fast, at least.

Our various build outputs can easily take up dozens of gigabytes per workspace. Multiply that across dozens of engineers, and it's considerable. We've been seeing a lot of disk utilization warnings lately because some of our disks are getting "too" full (even though in reality they often have hundreds of gigabytes of spare capacity). This is an easy way to quiet those warnings, and leave us with plenty of capacity to onboard additional engineers without needing to upgrade our actual storage resources.

Also bumps threshold to 90%, because realistically that's still aton of spare capacity.