Fix all remaining negative zero bugs

[GarboMuffin]

My understanding is that the new compiler tries to reason about -0 a good bit but the use of (x||0) prevents it from fixing all of the bugs. This replaces that syntax with a jsexecute function that returns 0 only for NaN and leaves -0 unchanged.

Draft because:

• Need more performance testing. Brief initial testing found no slowdown in Linux in Scratch. @Geotale said this might even be faster in ideal conditions.
• Need some solid test cases to add.

Test @ https://experiments.turbowarp.org/fix-negative-zero/

[Tacodiva]

How does this effect performance?

[GarboMuffin]

My only datapoint so far is that isNaN(value) ? 0 : value made Linux in Scratch run 50% slower in Firefox whereas value === value ? value : 0 made it have no measurable difference compared to develop. Maybe next weekend or so I can look around a bit more. I'm particularly worried about performance of generator functions in Firefox which don't get into the highest tier of JIT so probably no inlining at all in that case.

[GarboMuffin]

some years ago i thought about gathering up some hundreds of projects so i can make a proper performance tracking and comparison system, though i never got around to that. would've been really cool to have these last weeks

[GarboMuffin]

it's also on https://experiments.turbowarp.org/fix-negative-zero/ if you were interested in testing anything yourself

Linux in Scratch, in Firefox, with warp timer, goes from around 36s -> around 40s to boot so it's measurable though less catastrophic than I expected. Chrome in the same configuration is maybe just a tiny bit slower but I think it's within noise.

[Geotale]

I'm a tiny bit sad to see even any performance deficit on any browser (of course it would be Firefox tbh), but I'm extremely happy it's minor enough (and prooobably only in Firefox maybe? Idk I'd have to ask my friend who knows way too much about Chromium) this may be merged <3

[GarboMuffin]

the justification is

• firefox is used by like 1% of users (this applies to the internet broadly, but especially to us due to high usage of chromebooks in education environments)
• the performance loss is only when the function is compiled as a generator function, in which case Firefox is already 3-4x slower than the competition so this doesn't move that needle much
• most important parts of projects are compiled to regular functions, not generators, unless you enable warp timer in which case you are making a performance tradeoff anyways, so again this isn't moving the needle much