[ci] Refactor CI of test cases in integration module

[weiqingy]

Linked issue: #161

Purpose of change

Splits the Python unit-test job from the integration-test job using a new @pytest.mark.integration marker.

Today, 10 tests under python/flink_agents/integrations/*/tests/ need live external services (Ollama, DashScope, OpenAI, Azure, Anthropic, Chroma) or optional Python deps (mem0), but they live in the unit-test job and silently skip in 5 of 6 matrix cells — only the Python 3.10 cell auto-pulls Ollama. When Ollama regresses, exactly one cell reports failure, which looks like environmental noise.

After this PR:

• python/pyproject.toml — new [tool.pytest.ini_options] block registers the integration marker and enables strict_markers = true.
• 10 test files under python/flink_agents/integrations/*/tests/ — module-level pytestmark = pytest.mark.integration (list form for test_mem0_vector_store.py to preserve its existing skipif).
• tools/ut.sh — both if $run_e2e; then ... else ... fi blocks (uv branch + pip fallback) updated:
  • ut-python (run_e2e=false): appends -m "not integration" to the existing -k "not e2e_tests" filter. Drops 70 tests from the unit-test job (was 518, now 448).
  • it-python (run_e2e=true): runs two sequential pytest invocations — existing -k "e2e_tests_integration" (27 tests, unchanged) and new -m "integration" (70 tests) — with aggregated exit codes and an exit-5 trap on the integration arm to defend against -m selector typos that --strict-markers cannot catch.

No .github/workflows/ci.yml change required.

integrations/mcp/tests/test_mcp.py is intentionally not tagged: it self-hosts an MCP server via multiprocessing and runs deterministically without external deps.

Tests

Locally verified empirical contract (pytest --collect-only -q):

Selector	Tests	Notes
pytest -m "not integration" -k "not e2e_tests"	448	New ut-python arm (was 518)
pytest -m "integration"	70	New it-python arm 2
pytest -k "e2e_tests_integration"	27	it-python arm 1, unchanged
no selector	552	Unchanged

Overlap between -m "integration" and -k "e2e_tests_integration" is 0 — disjoint by directory structure.

Additional verification:

• ./tools/lint.sh -c passes
• ./tools/check-license.sh passes
• ./tools/ut.sh -p passes locally (448 passed, 104 deselected, ~28s)
• pytest --markers shows the registered integration marker with description
• Typo'd @pytest.mark.intergration triggers a collection error (strict_markers = true enforcement is active)

API

No public API changes. This PR is CI plumbing plus a pytest marker added to existing test files.

Documentation

• doc-needed
• doc-not-needed
• doc-included