Log message:
textprox/py-rapidfuzz: Exclude i386 from avx2 PLIST conditional

This package had untested code to expect three files that are
avx2-only when building on i386, but those files don't actually get
built.  This is almost certainly because not all i486-and-up CPUs have
avx2 instructions.

Resolves failure to package on NetBSD 9 i386 using
  cpu0: "Intel(R) Core(TM)2 Duo CPU     T8100  @ 2.10GHz"

Log message:
*: restrict py-numpy users to 3.9+ in preparation for update

Log message:
textproc/py-rapidfuzz: Spell amd64 correctly as x86_64

This is MACHINE_ARCH, not MACHINE.

Log message:
py-rapidfuzz: fix on x86

Log message:
Fix textproc/py-rapidfuzz on !x86

Log message:
py-rapidfuzz: Remove bogus compiler flags on SunOS.

Log message:
py-rapidfuzz: try fixing build with Python 3.8 and 3.9

Fix pkglint while here.

Log message:
textproc/py-rapidfuzz: Actually change to gcc10

(The previous commit explained why I chose 10 for GCC_REQD, but
accidentally contained 8 because I was  testing that.  While 8 seems
to work, various code is increasingly troubled by non-recent
compilers, and it doesn't seem useful to try to live on the edge.)

Log message:
textproc/py-rapidjson: Increase compiler requirements

On netbsd-9, some files failed to build with an ICE.  (The upstream
build system unhelpfully retried the build without the C extension,
making this harder to debug, but PLIST caught it.)

Declare C++17 because upstream README says so.
Require gcc 10, because gcc 7 on NetBSD 9 fails, and 10 is the new
normal.

Log message:
py-rapidfuzz: update to 3.1.1.

Changed

    upgrade to taskflow==3.6

Fixed

    replace usage of isnan with std::isnan which fixes the build on NetBSD