feat(ddi): columns the model doesn't lift ride along (#160)

[jstet]

The last of the #160 losses, apart from unregistered types (calculate), which stay out by design.

Every column the model doesn't read into its own fields now lives in columns on an item or choice. Only the XLSForm parser fills it; the lifted columns are listed in convention:ddiFields liftedColumns. Examples: media per language, choice_filter, read_only, Kobo's $kuid. The codebook carries them as follows:

where	note	subject
a question's var / varGrp, a group's varGrp	cdl:column	the column, e.g. media::image::Deutsch (de)
the first question using the list	cdl:choice_column	<code> <column>
stdyDscr, for note rows and rows without data	cdl:row_column	<row> <column>

ddi2xlsform writes them back. LimeSurvey is unchanged.

Schematron: the "one note per type and language" rule on var / varGrp now counts per subject, so several cdl:column notes are allowed.

Left, per ddi2xlsform/README.md:

• unregistered rows and empty groups
• settings that are neither a string nor a number
• columns of end_group rows
• appearance case
• whitespace in the type cell

Tests

• New unit tests cover the note shapes and the way back.
• The generated forms now add media (per language), read_only, choice_filter and a choices filter column. All 5 properties pass at 2000 runs locally.
• vitest: 1469 passed
• pytest validation + codegen: 150 passed, 4 xfailed
• worker gradle test, npm run validate and the drift check pass

🤖 Generated with Claude Code