Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@tests/test_l10n.py`:
- Around line 55-73: The test repeatedly constructs loop-invariant objects
inside the loop: the set mandatory_fields and the regex placeholder_re are
constant across iterations; move their creation out of the loop to avoid
recomputing them each time. Locate the for-loop that iterates over po files in
tests/test_l10n.py, hoist the mandatory_fields set (currently defined as
"Content-Transfer-Encoding", "Content-Type", etc.) to before the loop, and
likewise compile placeholder_re (the regex used around line 104) once before the
loop, then reuse these variables inside the loop and update any references to
them.
- Around line 68-72: The check building missing_fields should be simplified and
made deterministic: replace set(po_file.metadata.keys()) with
set(po_file.metadata) (or just po_file.metadata) when computing missing_fields =
mandatory_fields - set(...), and sort the missing field names in the assertion
message (e.g., use ', '.join(sorted(missing_fields))) so the failure message for
po_path is reproducible; update references around missing_fields,
mandatory_fields, po_file.metadata and po_path accordingly.

Verify each finding against the current code and only fix it if needed.

Duplicate comments:
In `@tests/test_l10n.py`:
- Around line 70-75: In tests/test_l10n.py fix the two nits: construct
missing_fields using set(po_file.metadata) instead of
set(po_file.metadata.keys()) and make the failure message deterministic by
joining sorted(missing_fields) (e.g., f"{', '.join(sorted(missing_fields))}");
reference the variables/members mandatory_fields, missing_fields,
po_file.metadata and po_path when applying the change.