Update CoMapeo fetch script to use new endpoints, including presetRef

[rudokemper]

Goal

Closes #174.

Before syncing this, we will need to do https://github.com/ConservationMetrics/gc-programs/issues/114.

What I changed and why

• Updated SAMPLE_OBSERVATIONS, and added SAMPLE_PRESETS for TDD; these were retrieved from a real CoMapeo cloud server running the latest version.
• Use the latest endpoint routes
• Include metadata fields now returned by /observation route
• For each observation, fetch and extract preset data using a new helper function fetch_preset. This enables us to finally add, among other things, the category name, which has been requested by virtually all of our users. (Another nice one: the hash for color, so we can use the same colors e.g. in GC Explorer as what users saw on their CoMapeo map for each category)

What I'm not doing here

• [comapeo] Support downloading tracks in addition to observations #101
• Downloading all of the presetRef data for project safeguarding [comapeo] Download all presetRef data to have a complete record of the entire config #178
• Fixing a flaky test for GFW alerts causing CI to fail here (and reproduced locally). [gfw] fix flaky max_months_lookback test #176
• Download the referenced icon via /icon/:docId. Perhaps that can be part of [comapeo] Download all presetRef data to have a complete record of the entire config #178

LLM use disclosure

Tests generated and improved by Claude Sonnet 4.5 in Cursor, with a strong amount of hand-holding.

[IamJeffG]

Here and presets, is there a reason we use singular and not plural?

The reason to (in theory) use plural is to allow a hypothetical endpoint that lists all observations (or all presets):

/projects/abc123/observations
/projects/abc123/presets

Also for consistency with "projects" in the same URL.

But I will not press this point if you prefer to leave it.

[IamJeffG]

Oh, this is not our code, this is defined upstream isnt it?

[rudokemper]

Correct. See digidem/comapeo-cloud#61 (review)

They have changed some of the routes to singular, but not all (though it is their intention).

[IamJeffG]

it's not a big deal yet so maybe we can sleep on this suggestion for now...

A common pattern is to not pass nitty-gritty details for the request like access_token around to different functions, but instead to instantiate a requests.Session() one time and pass that around.

So higher up in the stack (transform_comapeo_observations or even whatever calls it) would do this:

session = requests.Session()
session.headers.update({
    "Authorization": f"Bearer {access_token}"
})

# ... 

fetch_preset(..., session)

And then here:

def fetch_preset(server_url, session, project_id, preset_doc_id):
    try: 
        response = session.get(url)
        response.raise_for_status()
        # etc
    except ...

[rudokemper]

Great suggestion, thank you. I implemented this.