fix(#1977): retire the iccPawgReport --read option

[colourbill-ctrl]

Part of #1977.

Retires iccPawgReport --read. @xsscx confirmed on #1972 that it was leftover prototyping scaffolding rather than an intended feature, and #1977 records the mechanism and scope.

Why it can never have worked

DumpPawgReport() only consulted the flag after the strict path had already returned NULL:

  CIccProfile *pIcc = ValidateIccProfile(szFilename, sReport, nStatus);

  if (bUseRead && !pIcc) {
    pIcc = ReadIccProfile(szFilename);
  }

ValidateIccProfile() returns NULL on a parsed profile only when ReadValidate() reaches icValidateCriticalError, and the two conditions that get it there are exactly the conditions CIccProfile::Read() — behind ReadIccProfile() — bails on:

condition	ReadValidate()	CIccProfile::Read()
ReadBasic() fails	critical, IccProfile.cpp:941	return false, :874
any LoadTag() fails	critical, :992	return false, :905
CheckTagLayout()	tops out at icValidateWarning	not consulted

Supporting facts, both checked rather than assumed: icValidateStatus is ordered OK(0) < Warning(1) < NonCompliant(2) < CriticalError(3) (IccDefs.h:106-111), so >= icValidateCriticalError matches only the maximum and the NonCompliant/Warning header checks cannot trigger the NULL return; and LoadTag() opens with an absolute Seek(offset, icSeekSet), so the repositioning ReadValidate() does beforehand cannot change its outcome relative to Read()'s loop.

The root of the mistake is that ReadIccProfile() is the eager loader, not a permissive one. On tag structure it is stricter than a lazy OpenIccProfile(), so it fails on precisely the conditions that just made ReadValidate() report critical.

Nothing is lost

The graceful degradation --read was reaching for already exists and never depended on it. LoadRawProfile() runs unconditionally at the top of DumpPawgReport(), before and independently of any IccProfLib parsing, so a file the library refuses still receives the full raw-byte security assessment with the parsed-profile items reported as NOT RUN. That reasoning is now recorded where the fallback used to sit, because it is what makes a strict-only load acceptable here.

Retired, not accepted-and-ignored

--read now falls through to the existing unrecognised-switch branch:

$ iccPawgReport --read profile.icc
Unknown option "--read"

Usage: iccPawgReport {--json} profile
...
$ echo $?
1

A caller still passing it is told plainly, rather than believing a fallback ran.

Scripts move with the source

icc-pawg-qa-scan.sh drove two of its four QA variants through --read and runs in CI from ci-iccdev-tool-tests.yml. Leaving it would turn those two into exit 1. They are dropped rather than replaced — both executed the same code path as their non---read counterparts.

pawg-read-option-fidelity, which only ever asserted that a good profile reports identically with the flag set, is replaced by pawg-read-option-retired, which pins the rejection.

Not touched

iccDumpProfile has a separate --read with different semantics — eager ReadIccProfile() instead of lazy OpenIccProfile(), not a post-failure fallback (iccDumpProfile.cpp:210). It is live and working, and along with icc-dumpprofile-qa-scan.sh and docs/ctest.md:273 is deliberately left alone.

Verification

Acceptance test from #1977 — report output unchanged. All 80 tracked Testing/*.icc profiles, text and JSON modes, version string normalised: 12639 lines, 0 diff against pre-removal master.

The new regression case is red-tested. Silently re-accepting --read makes it fail with --read was accepted (expected non-zero exit; the option is retired), so it catches the accepted-and-ignored mode specifically rather than just the presence of the flag.

Build and suite.

• IccPawgReport.cpp and PawgReport.cpp compile clean under -Wall -Wextra -Wpedantic -Werror on gcc and clang
• shellcheck -S error clean on both changed scripts
• 7 PAWG CTests pass
• full local suite 148/150; the two failures are environmental and reproduce independently of this branch — spectral-tiff-preview (missing python imagecodecs) and qa-profile-manifest (0 mismatched, one unlisted stray untracked file in my working tree)

Rebased onto c0cd71db so CI validates the landing tree.