[doha-230]

Both harnesses could report a broken build as a success, and one of them could hang until the CI job timeout:

• build_and_run_all_cli_tests: the status of ../bin/$t was lost in "../bin/$t | sed ... | grep -v 'Using QtTest'": with "set -e" the status of a pipeline is the status of its last command, so a test binary that failed was reported as a success as long as grep printed a line. The status is now read before the coloring pipeline, every binary is bounded by ENGAUGE_CLI_TEST_TIMEOUT (default 300s), and the script exits non-zero when any test failed or timed out.
• build_and_run_all_gui_tests: a FAIL was only printed, and the script exited 0 in every case, so a GUI regression could never fail the build. It now counts the failures and timeouts, prints a summary, and exits non-zero. Every case is bounded by ENGAUGE_GUI_TEST_TIMEOUT (default 120s), because a hanging Engauge process otherwise consumes the whole CI job. A hang is reported as TIMEOUT with the case name instead.
• build_and_run_all_gui_tests: cases that download an image from a web site, such as drag_and_drop_http, depend on an external site and cannot run without internet access. ENGAUGE_SKIP_NETWORK_TESTS=1 skips them (opt-in; the default still runs every case).
• ci-test.yml: qt6-image-formats-plugins was missing, so GIF/JPEG/TIFF loading fails. Engauge then shows a modal warning dialog and waits for input forever, which is what made the drag_and_drop regression cases hang.

Verified with a stub Engauge and a stub test binary: a failing binary now gives "FAIL : (exit status 1)" and exit status 1 (the old script printed PASS and exited 0), a hanging binary gives "TIMEOUT: " and exit status 1 instead of blocking, a passing binary still exits 0, and ENGAUGE_SKIP_NETWORK_TESTS=1 skips only the http cases.