Improve test suite performance via module-scoped fixtures and batched…

[ehelms]

Why are you introducing these changes? (Problem description, related links)

Test runs were spending a large fraction of their wall-clock time recreating
the same remote resources for every test. The ten Foreman/Katello resource
fixtures (organizations, products, repositories, lifecycle environments, content
views, activation keys) were all function-scoped by default, so each individual
test paid the full create-over-API cost. Additionally, two pulp-manager tests
each spawned a separate Django process inside the container, doubling that
startup cost, and inherently-slow tests (backup, target lifecycle) were mixed
with fast ones making it impossible to exclude them during development.

Fixture call reduction (measured with --fixture-durations):

Fixture	Before calls	Before total	After calls	After total	Saved
Pulp manager Django startup¹	2	87.5s	1	24.3s	63.2s
yum_repository	4	61.2s	2	10.8s	50.4s
client_environment	2	35.2s	1	13.6s	21.6s
organization	10	23.1s	3	7.9s	15.2s
product	7	3.5s	2	1.1s	2.4s
Total		210.5s		57.7s	152.8s

¹ Before, each test launched its own pulpcore-manager shell process (45.0s + 42.5s in test call time); after, both tests share one module-scoped fixture.

Test phase overhead (pytest --durations grand totals):

Phase	Before	After	Saved
Test setup	156s	2s	154s
Test teardown	120s	<1s	~120s

Skippable slow tests (with -m "not slow"):

Test module	Duration
backup_test.py	~2m 1s
target_lifecycle_test.py	~2m 46s
Skippable total	~4m 47s

What are the changes introduced in this pull request?

• tests/conftest.py — Promote all remote resource fixtures to
  scope="module" so resources are created once per test module, not once per
  test. Expand content_view teardown to handle version cleanup (previously
  left to test bodies, which broke with shared fixtures). Simplify
  client_environment teardown to delegate version cleanup to content_view.
  Change wait_for_metadata_generate to filter with state != stopped instead
  of fetching all historical tasks. Register the slow marker in
  pytest_configure.
• tests/pulp_test.py — Add _PULP_MANAGER_SCRIPT constant and module-scoped
  pulp_manager_info fixture that runs one Django startup via
  pulpcore-manager shell (base64-encoded to avoid shell quoting issues).
  test_pulp_manager_check and test_pulp_import_export_settings now consume
  the shared result instead of each launching their own container process.
• tests/feature/katello/api_test.py — Remove per-test content view version
  cleanup loop from test_foreman_content_view; the content_view fixture
  teardown now handles this.
• tests/backup_test.py — Add pytestmark = pytest.mark.slow.
• tests/target_lifecycle_test.py — Add pytestmark = pytest.mark.slow.

How to test this pull request

Steps to reproduce:

• Deploy a foremanctl environment and run the full test suite:

  python -m pytest tests/ --durations=20 --fixture-durations=10 -v
  

• Verify resource fixtures appear once per module in the setup column, not once
  per test
• Deselect slow tests and confirm the run completes faster:

  python -m pytest tests/ -m "not slow"
  

• Confirm the content_view teardown cleans up published versions without
  error when the katello feature is enabled

Checklist

• Tests added/updated (if applicable)
• Documentation updated (if applicable)

[ehelms]

There's got to be a better / cleaner way to do this, consider this just a "hack" for now.