CI note from the first run:

• The CPython failures are in test_draft_docs_with_comments_from_user_guide, with other_unpublished.html unexpectedly built / missing the preview log assertion. That is the draft-docs fixture indentation issue fixed separately in Use PyPy 3.11 in CI #4121.
• The PyPy 3.9 Ubuntu/Windows failures are the current Hatch dependency bootstrap issue (cryptography 48 / PyO3 requiring PyPy 3.11), also covered by Use PyPy 3.11 in CI #4121.

I am keeping this branch scoped to the missing static-file source behavior instead of copying those CI-only fixes into this PR.

Follow-up after the rerun: the code/test failures from the first matrix are gone. The only remaining red checks are both PyPy 3.9 dependency bootstrap failures before MkDocs tests run:

• test (pypy-3.9-v7.x, ubuntu-latest): python -m pip install --upgrade hatch falls into building cryptography from source and fails in the PyO3 build (pyo3-ffi).
• test (pypy-3.9-v7.x, windows-latest): the same hatch install path falls into building zstandard from source and fails through CFFI/distutils (No module named 'distutils.msvc9compiler').

All CPython, macOS PyPy, lint, style, build, and package jobs are green on this head. This matches the PyPy 3.9 CI bootstrap drift addressed separately by #4120 rather than a failure in this static-file change.

I pushed a small follow-up (e99222bf) to dedent the existing draft_docs user-guide fixture. The first CI run exposed the same pathspec 1.x parsing issue seen on the PyPy CI PR; after the follow-up, the CPython and macOS PyPy test lanes that reach the test suite are green.

Fresh local validation on the current branch:

• TMPDIR=$PWD/.tmp-test .venv/bin/python -m unittest mkdocs.tests.build_tests.BuildTests.test_draft_docs_with_comments_from_user_guide mkdocs.tests.structure.file_tests.TestFiles.test_copy_file_missing_source mkdocs.tests.structure.file_tests.TestFiles.test_copy_file_dirty_missing_source -v -> passed
• TMPDIR=$PWD/.tmp-test .venv/bin/python -m unittest mkdocs.tests.structure.file_tests -v -> passed (34 tests, OK (skipped=1))
• TMPDIR=$PWD/.tmp-test .venv/bin/python -m unittest discover -s mkdocs -p '*tests.py' -> passed (727 tests, OK (skipped=4))
• uvx --with pip hatch run style:check -> passed
• git diff --check origin/master...HEAD -> clean

Current GitHub Actions after the follow-up: build/lint/package/style and all CPython test lanes are passing. The remaining red lanes are the existing PyPy 3.9 dependency-install failures, before tests run: Ubuntu fails building cryptography because PyO3 now requires PyPy >= 3.11, and Windows fails building zstandard through the hosted PyPy 3.9/CFFI distutils path. Those are unrelated to this missing-static-source fix and are covered by #4120.

CI note: the two red PyPy 3.9 jobs are failing before the MkDocs test suite starts, during the workflow dependency install step (python -m pip install --upgrade hatch).

• pypy-3.9-v7.x, ubuntu-latest: Hatch dependency resolution reaches cryptography, which builds from source and fails in pyo3-ffi with the configured PyPy interpreter version (3.9) is lower than PyO3's minimum supported version (3.11).
• pypy-3.9-v7.x, windows-latest: Hatch dependency resolution reaches zstandard, which builds from source and fails in CFFI with ModuleNotFoundError: No module named 'distutils.msvc9compiler' / This CFFI feature requires distutils.\n\nThe macOS PyPy job and all CPython test jobs passed, so I do not see evidence that this branch changed MkDocs behavior on PyPy; this looks like dependency/tooling drift in the PyPy 3.9 CI environment.