Fix test runner to use current Python and report failures

[xdong99]

Summary

• Run test scripts with sys.executable so the active virtualenv/CI Python is preserved.
• Resolve test files relative to the tests/ directory, allowing python tests/runtests.py from the repo root.
• Exit non-zero when any test script fails.
• Replace os.system("python -m docxtpl ...") with subprocess.run(..., check=True).

Verification

• python tests/runtests.py
• python -m flake8 docxtpl tests setup.py --count --max-line-length=127 --show-source --statistics

[arturict]

Good docs-oriented PR. Could you include a short negative test case for the new placeholder support to ensure error behavior remains stable?

[JackSpiece]

Checked this locally from the repository root with the project environment. All 36 test scripts completed and the runner returned 0. I also reran it with the project import deliberately unavailable; it collected the failing script names and returned 1 as intended. The module CLI test used the same interpreter and generated its output successfully. The full flake8 target passes too. No blocking issues found.

[yangfan-yf-yf]

The new path handling still breaks the documented invocation on supported Python 3.7 and 3.8. Before Python 3.9, __main__.__file__ retains the relative script path; Python 3.9 changed it to an absolute path (Python 3.9 notes). With cd tests && python runtests.py, os.path.dirname(__file__) is therefore "", and the first subprocess.run(..., cwd=tests_dir) raises instead of recording a test result.

I reproduced the relative-__file__ branch on 9721512 with runpy.run_path("runtests.py", run_name="__main__"): it prints cellbg.py ... and then raises OSError: [WinError 123] at the first child launch. The repository documentation still gives cd tests/; python runtests.py, and the package still advertises Python 3.7 and 3.8.

Could tests_dir be normalized to an absolute path first (for example, os.path.dirname(os.path.abspath(__file__))) and the documented invocation be covered on one of the oldest supported versions? The root invocation otherwise ran all 36 test scripts successfully, and a synthetic failing script was aggregated and returned exit 1 as intended.