Test: guard the repository paths README.md tells users to download

[AmaadMartin]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

2. Or, if no issue exists, describe the change:

Problem: README.md tells users to curl four paths out of this repository. Nothing asserts that those paths exist; for a while one of them did not, and the test suite stayed green through the whole window. scripts/update_constraints.sh and the pyproject.toml classifiers can also drift apart, which leaves a supported interpreter with no published constraints file.

Solution: One new test module reads the real README.md, the real scripts/update_constraints.sh, the pyproject.toml classifiers and the five committed constraints files. It fails when a documented path is absent from the checkout, when the two interpreter lists disagree, when a constraints file is missing, empty or no longer pins exactly, or when the two-line header stops carrying the --exclude-newer date that update_constraints.sh --check reads back out of it. A last test asserts the README still documents the recipe, so the guard cannot be defused by deleting its subject. No production file changes: the diff is one added file.

Collision check: I compared this against every open pull request on the fork. Four are adjacent. #108 adds a README link guard, but it also deletes the install recipe on the premise that no constraints file was ever published; that premise no longer holds, so I did not stack on it. #159 pins how the files are generated, #149 the generator CLI, #163 the pre-commit hook. None of them guards that a documented download exists.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

pytest tests/unittests/test_constraints_files.py -q gives 27 passed.

Every guard was run against a mutated tree and confirmed to fail:

Mutation	Failing test
mv constraints-3.10.txt /tmp/	test_readme_repo_references_resolve[constraints-3.10.txt], test_constraints_file_exists_for_every_supported_version[3.10]
Append a raw.githubusercontent.com/.../nope.txt link to README.md	test_readme_repo_references_resolve[nope.txt]
PYTHON_VERSIONS=() in the script	test_supported_versions_match_pyproject_classifiers
Delete the 3.14 classifier from pyproject.toml	test_supported_versions_match_pyproject_classifiers
Cut one space from the # uv pip compile header	test_constraints_header_is_machine_readable[3.11]
Change one a2a-sdk==1.1.1 pin to >=	test_constraints_files_pin_exact_versions[3.12]
Delete the install recipe from README.md	test_readme_documents_the_constraints_install_recipe, test_readme_reference_extraction_is_not_empty
Drop blob|tree from the URL pattern, so the CI badge leaks in	test_repo_paths_in_captures_only_paths_into_this_repository and 4 others

The third mutation is the one worth calling out. An empty PYTHON_VERSIONS array makes the parametrized tests collect zero cases, which pytest reports as skipped, not failed. test_supported_versions_match_pyproject_classifiers asserts the list is non-empty for exactly that reason.

Branch coverage of the new module is 95%, measured with pytest --cov --cov-branch. Two environment guards account for the shortfall. Lines 49-50 are the tomli fallback that Python 3.11+ never takes; CI runs the 3.10 leg, which does take it. Line 91 is the module-level pytest.skip for a tree that does not export the source checkout layout. I verified that path by moving scripts/update_constraints.sh aside and re-running: 1 skipped. I kept both rather than delete a guard to raise the number.

The module takes the repository root from the directory's shared isolated_import_utils.REPO_ROOT.

Manual End-to-End (E2E) Tests:

curl -fsS -o /tmp/constraints-3.10.txt \
  https://raw.githubusercontent.com/google/adk-python/main/constraints-3.10.txt
head -2 /tmp/constraints-3.10.txt
wc -l /tmp/constraints-3.10.txt
diff /tmp/constraints-3.10.txt constraints-3.10.txt

The download returns the two-line uv header, 1951 lines, and is identical to the committed file.

pytest tests/unittests --collect-only -q -k constraints_files

Collects the 29 new tests from the suite root, so the CI unit-test job runs them.

CI: Unit Tests, Mypy Check and A2A v0.3 Tests pass on every interpreter. Pre-commit Linter fails, and it fails for a reason this change does not touch: the update-constraints hook regenerates the files against a rolling "4 days ago" date and rewrites the --exclude-newer header, so it exits 1. The same job is red on main at 352d11d, the commit this branch sits on, and on the four main runs before it.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.