[survey2ddi 4/6] LimeSurvey RemoteControl adapter + formtransform limesurvey command

[jstet]

Step 4 of 6 for the survey2ddi retirement. Plan: HANDOVER_SURVEY2DDI.md. Depends on the normalize_responses port for its transform subcommand; the list / pull half is independent.

Why

limesurvey2ddi in CorrelAid/survey2ddi pulls a survey's structure and responses over LimeSurvey's RemoteControl API and converts them. The conversion is duplicated here; the pull is not.

What to port

limesurvey2ddi/client.py + lstsv.py:

• JSON-RPC over RemoteControl (/admin/remotecontrol): get_session_key → calls → release_session_key. RPC errors come back in the result body, not as HTTP status codes — the error shape and session expiry are the parts that need care, and the reason this is a bigger issue than the Kobo one.
• list_surveys(), get_responses(survey_id), and the survey-structure TSV export that feeds lstsvToDdiXml.
• pull(survey_id, out) writing the structure TSV + responses.

CLI:

formtransform limesurvey list
formtransform limesurvey pull <survey_id> -o output/
formtransform limesurvey transform <survey_id> -o output/   # pull + lstsv2ddi + data CSV

Env vars keep their current names so existing .env files work: LIME_SERVER_URL, LIME_USERNAME, LIME_PASSWORD, each overridable by a flag.

Prior art in this repo

tests/live/ already drives this exact API — through the citric Python client, via codegen.limesurvey_stack (session handling, wait_until_ready, import+activate). Read that for the call sequence before writing the RPC layer.

Worth deciding as part of this issue: once a TS client exists, should the live suite exercise it instead of citric? That would make the client's session handling continuously tested against a real LimeSurvey rather than only in unit tests with mocked responses. Not required to close this, but say which way you went and why.

Constraints

• Node-only; not re-exported from src/index.ts (same rule as the Kobo adapter).
• Never log credentials or the session key.
• Always release the session key, including on error paths.

Acceptance criteria

• Session acquire/release around every call, released on failure too
• RPC-level errors surface as real errors, not silent nulls
• list / pull / transform match the Python CLI's behaviour and env vars
• Decision recorded on whether tests/live/ now uses the TS client
• Browser entry unchanged