reduce per-call overhead in RegisterFunc and callback dispatches

[tmc]

Operating System

• Windows
• macOS
• Linux
• FreeBSD
• Android
• iOS

What feature would you like to be added?

A reduction in overhad for purego calls

Why is this needed?

RegisterFunc/CFunc/15args costs ~1845 ns with 44 allocs. Profiling shows the hot spots are repeated reflect.StructOf in Darwin ARM64 stack bundling (rebuilds struct type every call), heap-allocated []reflect.Value in callbackWrap (38.5% of total allocs), deferred cleanup closures, and per-call reflect.Kind dispatch. All of this is internal — no API changes needed.

[MatejMagat305]

I am writing here, but it is question:
in normal case is it irrelevant, but similar to gen2brain/raylib-go#539 (this is weak examples, but I use it for explain principle what I means), it exist some case that I want call more than 1 function in one circle fps- Maybe make some special statement:

purego.Do(func(){
openGLCall1()
openGLCall2()
openGLCall3()
...
openGLCallN()
})

or

block := purego.Block()
block.Add(openGLCall1, args)
block.Add(openGLCall2, args2)
block.Add(openGLCall3, args3)
...
block.Add(openGLCallN, argsN)
block.Do()

where would be "prepare" cgo (thread, stack, ...) and call with them all in block ...., but it is true 1.26 reduce overhead and I do not know enough to do it..., therefore I ask, would it be possible (useful) do something like that?