[REVIEW] 16569 want a more foolproof method to build native tools

[REVIEW] 16569 want a more foolproof method to build native tools

[Richard Lowe]

https://code.illumos.org/c/illumos-gate/+/3524

16569 want a more foolproof method to build native tools
15552 NATIVE_LIBS needs a more truthful name
16570 tools svccfg could be built more normally

Unfortunately this is hard to split up in a way that still makes
logical sense to integrate, or at least I've failed to.

#15552 I put up for review long ago, and Gordon and Andy Stormont
looked at it, but I withdrew it because it opened up a can of worms.
This is the lid for the can of worms, via the arm64-gate where the
distinctions here are obviously far more important.

This is still not unfortunately _great_, but it is markedly better.

Something especially to draw your attention to is the addition of the
`--tag` parameter to cw(1ONBLD).  This does nothing at all, except by
its existence lets you annotate the command line for what it is doing
(in this case, either "native" or "target"), making it much easier to
audit for what's going on.  I want to draw your attention to this
because having a parameter that does nothing is, to say the least, a
weird design choice, and to point out that while this makes it
_possible_ to much more easily seek out classes of bugs where we use
the proto area or adjunct incorrectly, I haven't actually _done that
yet_.