kernel: remove SyOriginalArgv & SyOriginalArgc

[fingolfin]

These kept references to the argv array around. This is potentially problematic for consumers of libgap API, specifically when calling GAP_Initialize it is not obvious that one has to keep the argv passed to it alive until after GAP completed its startup.

While we could amend the documentation for GAP_Initialize (which unfortunately currently does exist at all) to mention this, it seems better to obviate the need for it.

This patch is a first step towards this. However, several pointers to argv members remain, at least these:

• SyCompileOutput
• SyCompileInput
• SyCompileName
• SyCompileMagic1
• SyRestoring

[ChrisJefferson]

Clearly moving things in a good direction

[fingolfin]

The Cross compiler for s390x CI job consistently fails, across multiple runs by now. So this doesn't look like a pure fluke anymore.

But I still can't figure out why it would do that. What change in this PR could even be responsible?!?

@ChrisJefferson if you have any idea at all...

[fingolfin]

It hangs in test suite testworkspace, which may give a clue...

[fingolfin]

OK, dev/ci.sh testworkspace also hangs on my mac! Now I have to wonder: why do all other CI tests pass?

[fingolfin]

So the problem was that when restoring a workspace, KernelArgs was reset to its value from when the workspace was created, not the current list of command line arguments.

The true mystery thus is why testworkspace still passed on Linux. Weird. In any case, this was a close one, only our cross compile test suite running on OpenBSD caught it. I now made sure we also run testworkspace on macOS where the issue also appears.

[limakzi]

@fingolfin

• Ad Continuous Integration.
  I suppose we could make builds on macOS as well. [1]

• Ad code. LGTM. 🎉

[fingolfin]

@limakzi we already run CI tests on macOS runners, just fewer than on Linux because they are bottlenecked (we can have 40 concurrent Linux jobs but only 5 concurrent macOS jobs, IIRC)

[fingolfin]

@limakzi thanks for the review!