Smaller "Faster println()" (partial revert of that brand-new addition) and precompiles

[PallHaraldsson]

Partial revert of #56308 and precompiles.

Would be nice to backport to 1.11.2 since printing common enough, and precompiles help it.

[PallHaraldsson]

@KristofferC this smaller revert likely better to backport to 1.11.2 (though I'm pretty sure I saw stack-allocation go away, not dreamt it... until I figure out how, leave code for it out).

Not addressing the small regression, unrelated to this: #56326 (comment)

[fingolfin]

This PR has no explanation as to what motivates it: what problem does it address respectively what does it enhance? Please clarify that, e.g. by editing the PR description.

[PallHaraldsson]

It's a partial revert of my merged PR that same day. I realized the extra code isn't needed so this is more like status quo of last week.

The extra code isn't a bug per se, but would require compilation, so better with precompiles or just not have it.

I also slipped in some good (IMHO) precompiles.

I think this can uncontroversially merged.

It passes CI (Windows must be false alarm like for musl):

Process failed to exit within 10800s, requesting termination and coredump (SIGQUIT) of PID 6540.