[SrkianFardin25]

Closes #1657

Problem

ProcessLock.acquire() in scripts/ingest_processor.py opened the lock file with open(path, 'w+'), which truncates it before flock is tried. So a second caller:

1. erased the owner's PID,
2. failed flock,
3. read an empty (non-digit) PID, decided the lock was stale and called _cleanup_stale_lock(), which unlinked the path while the owner still held its kernel lock.

A third caller then created a new inode at the same path and got the lock, so two ingest processes could run at the same time. release() also unlinked the path, which can delete a lock file that another process has just created.

The same "non-PID content means stale" check also fired while cps/admin.py holds this lock during a library restore, because it writes restore_calibre_db into the file instead of a PID.

Fix

flock is now the only source of truth, which matches how the kernel behaves: a flock is dropped when the holder's descriptors close, including when the process dies. A lock left by a dead process is never held, so no stale-lock cleanup is needed.

• Open the file with os.open(..., O_RDWR | O_CREAT) (no truncation) and write the PID only after flock succeeds.
• A contender that cannot get flock just waits and retries until the timeout. It never truncates or unlinks the file.
• release() clears the PID while it still holds the lock, unlocks, and leaves the file in place.
• Removed _check_stale_lock() and _cleanup_stale_lock(). Nothing else referenced them. _get_holding_pid() is unchanged and now reads a PID that a contender can no longer erase.

The lock path, name, timeout and return values are unchanged, so initialize_runtime() and the restore code in cps/admin.py work as before. The cwa-init service still deletes leftover lock files at container start.

Tests

Added tests/unit/test_ingest_process_lock.py (marked unit):

• test_failed_contender_keeps_owner_lock_file: owner acquires; a contender times out; the path still exists with the owner's inode and PID; a third caller also fails; after the owner releases, the third caller gets the same inode.
• test_contender_does_not_treat_non_pid_content_as_stale: a holder that writes restore_calibre_db (like cps/admin.py) keeps its file.
• test_unlocked_file_left_by_dead_process_is_reused: an unlocked file with a dead PID is taken over, and the PID is cleared on release.

All three fail on main (the path is unlinked) and pass with this change.

What I ran locally (Linux, Python 3.13, requirements.txt + requirements-dev.txt; python-ldap, Flask-SimpleLDAP, py7zr and faust-cchardet were left out because they need a C compiler my environment does not have):

PYTHONPATH=.:scripts pytest tests/unit/test_ingest_process_lock.py
  main:     3 failed
  this PR:  3 passed

PYTHONPATH=.:scripts pytest -m "smoke or unit" -n 2
  273 passed, 27 skipped, 1 failed
  The one failure is tests/smoke/test_smoke.py::TestBasicFunctionality::test_required_directories_exist,
  which asserts that /config exists. My environment has no /config directory, and the
  failure is unrelated to this change.

ruff check scripts/ingest_processor.py tests/unit/test_ingest_process_lock.py
  The new test file is clean. ingest_processor.py goes from 67 to 65 existing findings,
  and this change adds none.

I did not run the Docker integration tests (tests/docker, tests/integration). Those are left to CI.