Build architecture specific setjmp .S code more efficiently, reduce liklihood of Gnu ld error

[LeeTibbert]

A better and stronger way to fix #3916

• Consolidate the four architecture specific setjmp .S files into one. That means that the single file
  presented to the compiler will always contain meaningful code (or an unsupported OS). This
  does away with 'throw-away' compilations and empty .o files which take up hundreds of bytes.

• Ensure that the new setjmp.S file has a guarded section at the end to ensure that the compiler
  always generates a .note.GNU-stack section when using a Linux compiler with ELF output .

• I did_not check the various "unwind.S" files because I have never seen ld.bfd complain
  about them. They may be an issue there.

[LeeTibbert]

No changes to docs, yet doc build fails. Bad vibes & intermittent failures abound in CI tonight.
Time to end this sprint and check it by the light of day.

---

Next morning:

The build retry mechanism is wonderful. My many thanks to the implementer.

Woke up this morning and the build is all green, great! Life is good.

[natsukagami]

LGTM, it definitely is a bit more annoying to have to maintain a single Assembly file (of all languages ;) ) with code of all architectures... But it cuts down a lot of annoyances on the user side, so I'm for it. Thanks!

[LeeTibbert]

natsukagami

Thank you for the review & sign-off.

You are being kind/restrained using the word "annoying". It is a hack which not even an author could love.

Ugly as it is, I think the architecture tests within it have grown shaggy/borderline-un-intelligible. Perhaps
having them in one place will lead to some consolidation & purification.

Not to get too far off topic, how is your continuations work coming along?

I was doing some work the other day and began to think: I wonder if this is the kind of place were
continuations might apply.

Cheers,

L.