DDI round trip: close the documented losses, round trips through LimeSurvey

[jstet]

Follow-up to #155 / #154. `src/pipelines/ddi2xlsform/README.md` lists what a CDL codebook doesn't give back. Most of it can be carried, in standard DDI where it fits and a `cdl:` note otherwise (convention:ddiFields):

• List names: `cdl:list` on a select whose list isn't named after it (a grid member: after its grid).
• `or_other` shorthand: `cdl:or_other` on the pair's `varGrp type="other"` when the companion was synthesized.
• Note names: `cdl:note_names` next to a lead-in `preQTxt` (consecutive notes come back as separate rows).
• Intro/outro note position: `cdl:position` on `stdyDscr` (`in= after=`).
• Metadata rows (`start`, `deviceid`, …): `cdl:metadata` on `stdyDscr`, positioned the same way.
• Language names: `cdl:language` (tag → the form's column name), `default_language` as authored.
• Range bounds as authored (in `cdl:parameters`, `valrng` stays).
• All settings, not only `style` (`cdl:setting`).

Left as losses: a note's hint, rows no registry type covers (`calculate`, …), choice-sheet columns other than `exclusive`, a group without a label.

More round trips:

• LimeSurvey TSV → DDI → Instrument equals the TSV's Instrument.
• XLSForm → LimeSurvey → DDI → XLSForm equals XLSForm → LimeSurvey → XLSForm.
• XLSForm → DDI → XLSForm → LimeSurvey equals XLSForm → LimeSurvey.
• DDI → XLSForm → DDI is a fixpoint.
• The same on generated forms (fast-check).

[jstet]

Follow-ups merged: #163 (cells as authored, unlabelled groups, messages without a constraint, notes that blank lines can't split), #164 (every column the model doesn't lift: cdl:column / cdl:choice_column / cdl:row_column), #165 (empty groups, end_group cells, appearance case, every setting). Still lost: unregistered types (calculate, by design) and cell spellings (type-cell whitespace, begin group with a space, a boolean setting as text). Not released yet.