BigInt is slow on most targets (or at least tests are)

[kLabz]

See #12309

The overhead is not negligible, with total CI times being multiplied by 3-4 on some targets. There are a lot of tests involved, but seeing how hl and jvm are still fine with it, it should be possible to have them run in a decent amount of time on other targets as well.

[flashultra]

What are the acceptable times for the tests?

Here are the latest times :

1. With BigInt tests included [57m 12s]
   https://github.com/flashultra/haxe/actions/runs/16222217009
2. Without BigInt tests [30m 7s]
   https://github.com/flashultra/haxe/actions/runs/16225820364

The slowest tests are :
mac-test (lua) 30m 6s vs 6m 37s
mac-test (cpp) 43m 15s vs 21m 3s

[flashultra]

For the BigInt test there are three steps, here is a comparison of the results for Linux vs Mac (cpp)

First step:
(Linux cpp ) execution time: 9.927 seconds
vs
(Mac cpp ) execution time: 17.415 seconds

Second step :
(Linux cpp ) execution time: 84.258 seconds
vs
(Mac cpp ) execution time: 138.411 seconds

Third step:
(Linux cpp ) execution time: 12.346 seconds
vs
(Mac cpp ) execution time: 14.479 seconds

There is not a big differences in executuion time , but Mac cpp takes 34m 29s , and Linux is 10m 49s
(without bigInt test results are Mac cpp - 16m 56s , Lunux cpp - 8m 29s
So the problem is probably in the compile time, which I'm not sure how to optimize.

About Lua test , Linux use Lua 5.1 and I think Mac use Lua 5.2 (not sure if this matter at all)
Linux lua BigInt tests takes 19m 8s
Mac Lua - 29m 37s
without BigInt is:
Linux Lua 4m 35s
Mac Lua - 7m 54s

Maybe it can be optimized, I don't know how. It's probably best to stop the Lua tests?

[Simn]

I've disabled the BigInt tests on eval today because tests runtime went from 0.3 seconds to 50+ seconds on my machine. I haven't identified the actual cause yet.