Str of a QWord formats it signed

• Track P (the Str intrinsic and the write(Text) formatter chooser, both in parser.inc).
• Found 2026-08-20 by an FPC differential probe over unsigned edge values, written while chasing the mixed-signedness comparison bug.

Repro

var q: QWord; s: string;
begin
  q := 18446744073709551615;
  writeln(q);          { 18446744073709551615 — right }
  Str(q, s);
  writeln('[', s, ']');{ [-1] — same value, same program, four lines apart }
end.

Cause

Three code paths format an ordinal, and each spelled "is this unsigned?" differently:

tyUInt64 and tyNativeUInt are two of the four source spellings of an 8-byte unsigned (QWord, UInt64, NativeUInt, PtrUInt), so the middle row printed -1 for pu: PtrUInt while getting q: QWord right — the narrower kind of double case, where even the path that HAD the dispatch had only half of it.

Fix

All three use the stdout path's predicate. Str picks StrQWord for any unsigned ordinal (Char and Boolean keep their own formatters), write(Text) drops the = tyUInt64 test for not TypeSigned. A narrower unsigned is zero-extended and formats identically either way, so there is nothing a width test would add.

Neighbour, deliberately not touched

High(QWord) / Low(UInt64) / High(NativeUInt) / High(PtrUInt) are still REJECTED at compile time ("undefined variable"), the documented refusal from bug-p-high-low-reject-the-64-bit-type-aliases: the const evaluator carries Int64, and answering -1 would turn a refusal into a wrong value. Refiled as feature-p-const-evaluator-carries-unsigned-64-bit so the representation change is tracked rather than rediscovered.

Gate

test/test_str_of_unsigned.pas — 23 assertions, all fpc -O- -Mobjfpc 3.2.2's: the four 8-byte unsigned spellings, 2^63 exactly, narrower unsigned, all five signed types, widths on both sides of the dispatch, expressions rather than variables, and the same values through a Text file. The pinned binary scores 17/23. Wired into make test; tools/gate.sh quick GREEN, self-host fixedpoint byte-identical.