Read previous msgids (#|) from PO files and keep them on update - #1466

pybabel update --previous writes the old msgid of a fuzzy message as #| msgid ..., but read_po doesn't parse #| lines and stores them as user comments. The next update --previous writes them back as translator comments:

# first run                      # every later run
#, fuzzy                         # | msgid "Save the file"
#| msgid "Save the file"         #: app.py:2
msgid "Save this file"           #, fuzzy
msgstr "Datei speichern"         msgid "Save this file"
                                 msgstr "Datei speichern"

The # | msgid line then stays in the catalog as a translator comment even after the string is fixed. The same happens to #| lines written by GNU msgmerge --previous.

Two changes:

• read_po parses #| msgid / #| msgid_plural (and their continuation lines) into Message.previous_id. Message has no field for a previous msgctxt, so a #| msgctxt line is dropped instead of becoming a comment.
• Catalog.update keeps previous_id for a message that matches exactly but is still fuzzy, so a second run keeps the #| line, like msgmerge.
• write_po wrapped a long previous msgid with the comment wrapper, which splits it in the middle of the quoted strings (#| msgid "" "Save the file ... window so / #| " "nothing is lost"). It now writes one #| line per line of the normalized string, the way msgmerge does, so it reads back unchanged.

With both, three pybabel update --previous runs on the example above keep #| msgid "Save the file".

test_previous_msgid_roundtrip, test_long_previous_msgid_roundtrip and test_catalog_update_keeps_previous_id_of_fuzzy_message fail without the change. pytest tests/messages: the only failure is test_4_num_plurals_checkers, which fails the same way on master here (my locale data comes from the installed Babel 2.17.0, as the checkout has none).