make some static muts thread locals

[jupyterkat]

• makes static muts thread locals
• functions that call byond stuff now panic when called outside execution context
• expunge some unsafe code for safer versions
• interned strings made more sane
• swap out mpsc for flume
• changes some more hashmaps to use faster versions of hashes
• cargo update, cargo fmt ran
• sped up the debug server by alot

[willox]

are you saying the polling on mpsc for every instruction was a bigger bottleneck than the new thread locals?

[jupyterkat]

It’s probably the mutex existing for absolutely no reasons at all, getting locks is expensive when you get it literally every instruction.

I just switched mpsc to flume because it’s the better choice overall lol

[willox]

I guess the thread locals in auxtools are ok, could we swap back the debug server ones though? At least the one that is accessed for every single instruction

[jupyterkat]

Is there a specific reason why? Is it the constant initialization check?

[willox]

The instruction hook function currently almost doubles init time on tg (or something like that). It's easily a case where we should be keeping it as optimized as possible.

Thread locals don't just add an init check, they're effectively adding extra layers of indirection.

[willox]

If anything I want to increase the simplicity (post-compile) of our global accesses rather than turning them in to TLS or any other kind of variable with indirection