Fix flaky E2E test test_failing_tmt_test_case.cy.js

[pellecchialuigi]

Fixed e2e test test_failing_tmt_test_case
Extended tmt failing test timeout to 5 minutes

Co-authored by Cursor

Solves issue #310

[pellecchialuigi]

What this PR changes

This PR stabilizes the flaky E2E spec test_failing_tmt_test_case.cy.js (issue #310). The Cypress spec was brittle, but CI failures were mainly from nested TMT/Podman inside the API container: guests could not start, and when they failed the runner could not persist the log (NUL bytes from conmon), leaving the run stuck in running.

1. Harden the Cypress E2E spec (5cf753b)

Rewrote test_failing_tmt_test_case.cy.js to be less timing- and selector-sensitive: bind the user-files path from localStorage uId instead of hardcoding user 1; share openTestCaseResults() for navigation; raise suite timeouts (15s) and UI waits (20s); poll for fail up to ~2 minutes; fix cleanup to target the table row. FMF duration for the dummy failing test went from 1m to 5m.

2. Sanitize TMT output before persisting (80647bb)

Conmon can emit \x00 in stderr. PostgreSQL rejects NUL in text columns, so publish() failed and the run stayed running. sanitize_db_text() strips NUL; TestRunner.publish() sanitizes log and report before commit; TMT decodes stdout/stderr with errors="replace". Covered by new tests in api/test/test_testrun_publish.py.

3. Pin TMT default guest images (bbbe401)

Unpinned fedora / debian tags made nested provision flaky. Defaults are now fedora:39 and debian:12.

4. CI timeout for the demo environment (99267c3)

The Run demo environment step has timeout-minutes: 20 so a hung run_demo.sh fails the job instead of consuming the runner.

5. Nested Podman for TMT container provision (5283158)

The API image has no systemd, so nested guests fail with the default systemd cgroup manager and overlay-on-overlay storage. Ship cgroupfs + file events and vfs storage into the API image, start the API container with --cgroupns=host, and patch TMT to drop TTY (-id) and pass --cgroups=disabled.

Together this lets the failing-TMT E2E path provision a guest, persist a real result (including error logs), and wait long enough in Cypress for that result to show up.