Proposal: Bump required mingw-w64 version to 3.0 (from 1.0)

[vszakats]

I propose bumping the required minimum mingw-w64 version to 3.0 (2013-09-20).

1.0 and 2.0 were released in 2011. It seems unlikely that many people use them.
The oldest downloadable toolchain (that I know of) comes with 3.0. It means these
old versions aren't CI tested, and perhaps also seldom tested elsewhere. Issues
were never reported for them either. The last bugfix update for both 1.0 and 2.0
was released in 2015.

The benefit for curl is that it allows dropping some exceptions for the older versions.
2.0 doesn't fully support 64-bit file offsets, it doesn't have ADDRESS_FAMILY type,
and if_nametoaddress is missing from 1.0 Windows headers and implib.
2.0 and 1.0 also doesn't support __MINGW_PRINTF_FORMAT.

(The maintenance benefit is that one doesn't have to look up that far back in the
release tarballs when researching a CRT or Win32 API feature.)

[Andarwinux]

Maybe consider just bumping up to 11.0. I doubt anyone would build latest curl with such an old MinGW.

[vszakats]

Just spent a day (so far) on a fallout caused by a v5.0 header bug,
making me also contemplate bumping further.

curl makes use of v6.0 (2018-09-16) to avoid the bug above (in #18012)
There was a former fix for v7.0 (and older) versions in e512fbf #14703.
Some tricks may be deleted by requiring v7.0 (2019-11-11).
The next step up is v9.0 (2021-05-22) for ALPN support.

No gain beyond that, at the moment.

By stepping up to v6.0 we lose the dl-mingw, CM 6.4.0-i686 CI job.
By stepping up to v9.0, also dl-mingw, CM 7.3.0-x86_64.

Going v4.0 would allow dropping the workaround for missing
CERT_CHAIN_EXCLUSIVE_ENABLE_CA_FLAG macro.
Going v5.0 would allow dropping the workaround for missing
freopen_s() prototype.
Going v7.0 or above would prevent using all of the toolchains offered
via https://downloads.sourceforge.net/mingw-w64/. These are fairly
old, but they were once the canonical mingw-w64 tools and somebody
(like curl in CI) may still be using them. AppVeyor CI has these
pre-installed: https://www.appveyor.com/docs/windows-images-software/
v6.0 would still allow using the 8.1.0 and 7.3.0 releases.

Another limiting factor is mingw-w64 version shipping in Linux distros.
bookworm has v10, bullseye has v8, buster has v6, stretch has v5:
https://sources.debian.org/src/mingw-w64/

Considering the above, v6.0 may a reasonable next step IMO. Though
it'd prevent cross-building from Debian stretch (aka old-linux in CI).