Fixes #1266.

Root cause: normalize() checks whether each line of the string exceeds width, but it does not account for the keyword prefix (msgstr , msgid , msgctxt ) that the caller prepends on the same line. A 116-char string with width=120 escapes to 118 chars — under the limit — so it is returned as a single line. The combined output msgstr "..." is 125 chars, which exceeds width=120.

Fix: Add a first_line_offset parameter (default 0) to normalize(). When the escaped string plus first_line_offset would exceed width, the multi-line ""\n"..." form is used, placing the content on its own line where the keyword prefix is absent.

All call sites in generate_po are updated to pass first_line_offset=len(prefix) + len("msgstr ") (and equivalents for msgid, msgctxt, msgid_plural, msgstr[N]).

Before:

msgstr "Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industrystandard"
# ↑ 125 chars — exceeds -w 120

After:

msgstr ""
"Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industrystandard"
# ↑ msgstr line = 9 chars; content line = 118 chars — both within 120

Two regression tests added to tests/messages/test_pofile.py.

Used an LLM to help trace the root cause and draft the fix; all output verified manually.