Flatten KoboToolbox matrix and repeat-group responses

[rudokemper]

Goal

Closes #248

Note to the reviewer: +340 of the line difference is just fixtures and conftest stuff. The actual code changes are +67 lines, and +90 are tests.

What I changed and why

• Added flatten_kobotoolbox_submission() in the Kobo connector script and call it from transform_kobotoolbox_form_data() so repeat groups and matrix questions become wide {group}/{index}/{leaf} keys instead of one JSON cell. Existing reverse_properties_separated_by="/" then maps those to flat SQL columns. See README and docstrings for more info.

How I convinced myself this is right

• In terms of the "why do this", by working with one of our highest tier partners who wanted to analyze matrix results and was not able to work with a JSON blob dumped in one cell. We had never encountered matrix questions in a Kobo questionnaire before this.
• In terms of "how I know this works", besides tests, I tried it in production with real data. The result is as described in the README.

What I'm not doing here

• Apply this to ODK (also uses XLSForm), but I did leave clear TODOs to track that future need if it arises.

LLM use disclosure

As I was working with a partner in country and didn't have a lot of time, I used Claude Opus 4.7 in planning mode to scheme this work out, and had it implement it too. Then I made various edits to ensure the code and its documentation is legible to me. I screened the code carefully to make sure it is valid and maintainable, making a few improvements here and there.

[conservationtimothy]

Is it possible for Kobo submissions to contain deeply nested repeat groups (e.g. a repeat inside another repeat)? The current flattening logic only handles one level of slash-keyed rows, so I was wondering whether nested structures are something we need to account for, or whether Kobo's output shape guarantees this isn't needed.

Note: I saw the "if/when ODK flattening is required" TODO, so if this is more relevant to ODK rather than Kobo, feel free to ignore.

Not a blocker since I saw this working in practice, just curious about the expected payload shape.

[rudokemper]

Great catch. Thanks for pointing this out. I tried it out with a real Kobo form and discovered that yes, you can have repeats inside of repeats, and we needed to implement recursive flattening to handle this. 980e25f

[conservationtimothy]

Thanks for this

[conservationtimothy]

Approved but don't know if @IamJeffG might wanna give this the once over.

[rudokemper]

Thanks. When you have time, please check out the PR for the downstream branch this merges info also.

[IamJeffG]

No need for extra approval from me on this one! thanks.