Tests: the eyedropper image-sampling test no longer flakes (colour settings changed by parallel tests)

[bflatastic]

tests_eyedropper::clicking_an_image_samples_its_pixel failed at random in parallel CI runs (#008181 instead of #008080). This finds the cause and removes it.

• Cause: all engine unit tests run in one process, and two colour-management tests briefly changed the process-wide Color Settings: the working RGB space and black-point compensation. They held tests_colormgmt::GLOBAL, but every test that shows or renders a colour reads those settings without the lock, so for that moment colours went through the wrong profile.
• Fix:
  • The two settings-changing tests move to their own test binary (crates/engine/tests/color_settings.rs), so they run in a separate process. A lock couldn't fix this, because every colour reader would need it. No engine unit test now leaves the Color Settings off their defaults.
  • The eyedropper test compares the stored colour instead of its hex string, so it doesn't depend on the settings at all.
  • every_command_survives_empty_params toggles the process-wide proof view (view.proofColors, view.overprintPreview, view.separationsPreview), which other locked tests assert on, so it now takes the existing GLOBAL lock.
• No commands, parameters or model fields change.

Checks

Rebased onto 0038e00. cargo xtask ci passes all 7 steps (fmt, clippy, test, assets, brands, layers, wasm): 1,886 tests, 0 failed.

Stress runs, with 0 failures in each:

• the failure reproduced 1 time in 100 filtered runs before the fix;
• after the fix, 300 full engine runs with 32 threads;
• 300 filtered runs;
• four concurrent loops of 60 full runs with 64 threads.

🤖 Generated with Claude Code