libc: the wide printf family aborts on %qd, %D, %O, %U, unknown conversions and a trailing % that FreeBSD formats; fix branch ready #5293

The wide printf family (swprintf, vswprintf, and snwprintf_s for the plain formats) aborts the guest on four things FreeBSD 11.4 vfwprintf.c formats: the q length modifier (%qd), the %D, %O and %U conversions, an unknown conversion character (%y, %5%), and a format ending in a lone % (swprintf(.., L"100%") returns -1 where FreeBSD stops and returns 3). This is the wide counterpart of #3975, whose fix covers only the narrow FormatWindows path.

Fix branch: https://github.com/husseinkad/AnyPS5/tree/fix/libc-wide-printf-extensions (two commits on a recent main: the change and a follow-up from review; libc/src/FormattingWide.cpp, tests/GuestWideFormatting.cpp with a new CheckFreebsdConversions, and one docs/dev/debt/libc.md line). q means long long, D/O/U are the long forms, an unknown character prints itself with width and the - flag (a surrogate pair prints whole and counts as one character), and a trailing % ends the format. %m, %', $, %l5d, %lqd, %qqd, %lp, %qp and %hq still throw as before, and snwprintf_s keeps rejecting all of these.

Checks:

• The new cases fail on main (abort, "swprintf q length") and pass with the change; guest_wide_formatting, guest_formatting, guest_standard_stream and file_streams pass.
• About 3 million random cases per variant against the real FreeBSD 11.4 vfwprintf/vswprintf (with its gdtoa), including surrogate pairs and lone surrogates: 0 regressions (no format that matched FreeBSD on main stops matching); snwprintf_s is identical to main on 700k formats; none of the 10k random formats that worked on main changed output or return value; %D, %O, %U with h, hh, l, ll, j and z and the 100%, a%- and %%% cases match.
• Three independent reviews. The first two found that snwprintf_s had started accepting q and that a surrogate pair used as an unknown conversion character was split into invalid UTF-16; both are fixed in the second commit. check_conventions.py is clean and pr_overlap.py finds no overlap.

Not fixed here: %#.0O of 0 prints nothing where FreeBSD prints 0 (the existing %#.0o bug, which O inherits); %0% and %010% pad with blanks where FreeBSD zero-pads (listed in the debt entry together with the other unknown-conversion padding), as do %05ls and %05lc; %LD and %Ld, stacked modifiers such as %hz, and flags after the width are rejected; truncation returns -1 without setting EOVERFLOW. Only Windows was built and run.

AI-assisted: yes