test(ci): add CLI smoke and image format round-trip tests; run ctest in CI

[mcoliver]

Summary

First pass at #60: run the existing tests in CI, and add end-to-end tests of the shipped command-line tools and image formats. Also fixes an intermittent crash the new tests found.

Stacked on #63. This PR's base is fix/windows-deps-and-oiio-write, because the format round-trip test needs the OIIO writer fixes. Merge #63 first; GitHub will then retarget this PR to main.

Crash fix: TwkMediaLibrary::Library worker thread race

utvio (and anything else that creates a media library, including the GUI) aborted about 1 in 40 runs, more often under load, with:

boost::lock_error: mutex lock failed in pthread_mutex_lock: Invalid argument
  TwkMediaLibrary::Library::threadMain()

• Cause: m_taskThread was declared before m_taskMutex, m_taskCond, m_taskStop and m_taskQueue. Members are constructed in declaration order, so the worker thread started and locked a mutex that didn't exist yet.
• Second bug: threadMain() also reset m_taskStop = false. If the destructor ran before the thread started, that erased the stop request and join() hung forever.
• Fix: declare the thread last, remove the reset, and only join() a joinable thread.
• Result: 0 failures in 900 parallel utvio runs. Before the fix it crashed within about 80.

New tests (src/test/CliSmokeTest, standard Python only)

• cli-smoke: each bundled CLI (utvio, utvls, utvpkg, utvpush, utvshell, utvprof, py-interp, openutv-diagnostics, openutv-check-updates) must start, find its runtime dependencies, and exit with the expected output within a timeout.
  • On Windows it uses the .cmd wrappers, the way users run the tools.
  • Windows loader error dialogs are suppressed, so a missing DLL fails fast instead of hanging CI.
  • Qt tools run offscreen except on macOS.
  • This is the fix: utvpkg fails to find dependencies on Windows 11 #58 failure mode.
• image-format-roundtrip: every image format utvio -formats advertises as writable gets a generated test image written to it and read back. Colored blocks in three corners are checked for orientation, channels and values, with a looser tolerance for lossy codecs.
  • Also fails when a format is advertised as writable but can't be written. That is how fix: utvio fails to write PNG, DPX, WebP and QOI (blank "ERROR:", exit code 255) #59 presented.
  • Against the current release build it fails all 37 formats. Against this branch it passes all 32 testable formats (5 are skipped with reasons: gto, igto, bw, pbm, pgm).

Bugs found by running the tests in CI, fixed here

• utvio on GPU-less Macs (VMs, CI, render nodes): it required a hardware pixel format and exited, even for -version. It now falls back to Apple's software renderer with a warning. UTV_SOFTWARE_GL=1 forces the software renderer.
• utvio on Linux without an X display: segfaulted on an unchecked XOpenDisplay()/glXChooseVisual(). It now exits with a clear error. GLX context creation failures are reported too.
• Linux on Mesa: glGenFencesNV & co. were linked directly, so binaries failed to load without NVIDIA's libGL. They're now looked up at runtime.
• Existing tests replaced LD_LIBRARY_PATH via cmake -E env. They now prepend to it instead (--modify ...path_list_prepend).

Known issues found (tracked, not fixed here)

• fix(linux): utv* tools staged without launcher wrappers after rv→utv rename #66: Linux launch wrappers lost in the rv→utv rename. CI sets LD_LIBRARY_PATH as a workaround.
• fix(windows): HEIC/HEIF advertised as writable but fail; FITS doesn't read back #67: Windows HEIC/HEIF advertised as writable but fail; FITS doesn't read back. Marked xfail on Windows.
• fix(linux): binaries fail to load with Mesa libGL (direct GL extension symbols); utvio FBO fails under llvmpipe #68: Linux on Mesa. More GL extension symbols are linked directly (glBlitFramebufferEXT, …), and utvio's FBO fails under llvmpipe. Linux CI excludes the 4 affected tests, and utvio is xfail in cli-smoke.

CLI

• openutv-check-updates --help prints help. Previously any argument ran a live update check that could open a dialog.

CI

• ctest --output-on-failure now runs after the macOS and Linux builds in branch-build, dev-build and release. Linux uses QT_QPA_PLATFORM=offscreen.
• The Windows ctest step is now blocking (it had continue-on-error: true), with a timeout long enough for the round-trip test.
• Locally on macOS: ctest passes 120/120 in about 12 s.

CI status

Branch CI passes on macOS, Windows and Linux (run 36659762676). The known issues above are reported as xfail or excluded with links, so they stay visible without blocking.

Not in this PR (rest of #60)

UI automation, testing the installed/signed package on clean machines, and movie/audio format round-trips.

🤖 Generated with Claude Code