test: fix integration test coverage with single process

[wak327]

Proposed Commit Message

test: fix integration test coverage with single process

enable_coverage.py wraps the ExecStart=/usr line of each stage service
with `coverage run`. Since the single process optimization (#5489) the
stage services are shims whose ExecStart is `sh -c '... | nc ...'`, so
nothing got wrapped and coverage was silently no longer collected. On
releases whose packaging disables the optimization, the script instead
exits early because the network stage is still cloud-init.service.

Wrap whichever of cloud-init-main.service and the stage services run
cloud-init from /usr, and fail when none do, so that a future unit
change cannot silently disable coverage again.

Fixes GH-6769

Additional Context

Fixes #6769

The script has to handle two unit layouts:

• Upstream (single process): cloud-init-main.service runs ExecStart=/usr/bin/cloud-init --all-stages, and the stage services are nc shims. The current script exits 0 without wrapping anything, so coverage is never collected.
• Ubuntu 22.04/24.04 packaging (debian/patches/no-single-process.patch on ubuntu/jammy and ubuntu/noble): there is no cloud-init-main.service. Each stage service runs /usr/bin/cloud-init ... itself, and the network stage is cloud-init.service. The current script exits 1 because cloud-init-network.service does not exist, so the assertion in install_coverage() fails.

I checked the new script against unit files rendered from systemd/*.tmpl and against the no-single-process variants of them. It wraps only cloud-init-main.service in the first case and the four stage services in the second. With no matching unit, it fails. I also confirmed that a Type=notify unit whose ExecStart is wrapped in python -m coverage run --append still reaches READY and writes .coverage to its working directory (systemd-run --user --wait -p Type=notify ...). I have not run the full integration suite with coverage enabled because I don't have LXD available here.

enable_profile.py has the same problem, and _collect_profile() in tests/integration_tests/conftest.py still expects one stats file per stage. I'm happy to fix that in a follow-up so this PR stays focused on the reported issue.

Test Steps

CLOUD_INIT_INCLUDE_COVERAGE=true CLOUD_INIT_CLOUD_INIT_SOURCE=<path to cloud-init deb> \
  tox -e integration-tests -- tests/integration_tests/modules/test_write_files.py

/tmp/cloud_init_test_logs/last/html/index.html should now report coverage for cloudinit/. Repeat with CLOUD_INIT_OS_IMAGE=noble CLOUD_INIT_CLOUD_INIT_SOURCE=NONE for the no-single-process layout.

Merge type

• Squash merge using "Proposed Commit Message"
• Rebase and merge unique commits. Requires commit messages per-commit each referencing the pull request number (#<PR_NUM>)