Back when the multi-decl overhaul
(#657) switched
all unchanged multi-decl members from Decl::print to mkString, these
fixes were needed to prevent 3C's output from getting worse on some
unchanged multi-decl members in the regression tests. The fixes are no
longer needed for that reason, but Mike still asked me to submit them.
They do benefit some more complex cases that occur in the regression
tests, so the updates to the expected output of those tests provide test
coverage for the fixes.

I'm including this because the code fix was similar to a fix that was
originally in the multi-decl overhaul before the same fix was needed in
#702 and I was
aware of the issue from item 4 of
#703.

the changes even though they seem about as ad-hoc as the existing code.

https://correct-computation.slack.com/archives/G01GKGKHMFD/p1626879436256100?thread_ts=1626818406.255000&cid=G01GKGKHMFD

As I dug into the reasons why all the conditionals in the PR so far were
needed under the current design of the rest of mkString, I realized that
the design could be straightened out a bit and correctly handle one more
case involving checked pointers to fixed-size arrays. This doesn't
address the more difficult problems with wild pointers to fixed-size
arrays or with wild pointer levels being in reverse order in general
(see #703).

Any array levels still pending when we reach a function pointer should
become part of the function pointer declarator even if some text for an
outer checked pointer level has already been added to Ss. However, any
array levels that are outside the checked pointer level and have already
been committed to EndStrs should remain at the end of the entire type
and not get sucked into the function pointer declarator. To maintain
this distinction, we need to use a separate `FptrEndStrs` list for the
array levels in the function pointer declarator.

Also update a link to go to an existing issue John pointed out.