FIX SVM click-simulation tests failing on headless CI

[mkiani12]

Fixes the red FEZrs Tests run on main (run 32463258511).

What broke

FAILED tests/tools/svm/svm_calculator_test.py::test_nonsquare_training_samples_use_opencv_xy_as_numpy_yx
FAILED tests/tools/svm/svm_calculator_test.py::test_nonsquare_output_shape_and_pixel_order
RuntimeError: SVMCalculator needs training_samples when no display is available.

My fault, from the non-interactive SVM PR. That PR added a display check in front of the OpenCV window so a headless run raises a catchable error instead of Qt aborting the process. These two pre-existing tests drive the interactive path with every cv2 entry point mocked — but not display_is_available(). The guard runs before any of the mocked calls, so:

• workstation: DISPLAY is set → guard passes → mocked window runs → tests pass
• CI runner: no DISPLAY → guard raises → process() never reaches the simulated clicks → tests fail

I updated test_process_creates_rgb_stack for exactly this when I changed its ESC behaviour, but missed these two because they live in a separate helper further down the file and passed locally.

Fix

Force the display check on inside _run_process_with_clicks, matching what the ESC test already does. The helper mocks the entire GUI, so whether a real display exists is irrelevant to what these tests assert.

Test-only change. No library behaviour is affected, and the guard itself is unchanged.

Verification

Ran the full suite under env -u DISPLAY -u WAYLAND_DISPLAY, which reproduces the CI environment:

headless:      421 passed
with display:  421 passed

Previously headless gave 2 failures. Worth noting for future changes to this module: a plain local pytest run does not exercise the headless path, so anything touching the display guard should be checked with DISPLAY unset.