[BUG] Notebooks: markdown cell `id` is regenerated on every GET

[Okabe-Junya]

similar to #554

Describe the bug

pup notebooks get returns a different id for every markdown cell on each call, while graph cells (e.g. timeseries) keep stable ids. The same happens through the raw passthrough (pup api v1/notebooks/{id}), so the ids change in the API response itself, not in pup's client layer. same layering as #554.

This makes markdown cells unaddressable: any get -> modify -> update workflow that relies on cell ids cannot reference a markdown cell across two calls.

To Reproduce

Save as /tmp/nb.json:

{
  "data": {
    "type": "notebooks",
    "attributes": {
      "name": "markdown cell id stability test",
      "status": "published",
      "time": { "live_span": "1h" },
      "cells": [
        { "type": "notebook_cells", "attributes": { "definition": { "type": "markdown", "text": "# Section" } } },
        { "type": "notebook_cells", "attributes": { "definition": { "type": "timeseries", "requests": [ { "display_type": "line", "q": "avg:system.cpu.user{*}" } ] } } }
      ]
    }
  }
}

pup notebooks create --file /tmp/nb.json   # note the returned notebook id
for i in 1 2 3; do
  pup api v1/notebooks/$NB_ID --no-agent | jq -c '[.data.attributes.cells[] | {id, type: .attributes.definition.type}]'
done

Observed (no writes between the three GETs):

[{"id":"2l892wv8","type":"markdown"},{"id":"5k2rqfa5","type":"timeseries"}]
[{"id":"fwhey19c","type":"markdown"},{"id":"5k2rqfa5","type":"timeseries"}]
[{"id":"sjguoeel","type":"markdown"},{"id":"5k2rqfa5","type":"timeseries"}]

The markdown cell gets a fresh id on every GET; the timeseries id is stable.

Expected behavior

Cell ids should be stable across GETs of an unchanged notebook, and unique within a single response, for all cell types.

Additional observations (same root symptom, not minimally reproduced)

On a long-lived notebook that had received several notebooks update calls, GET responses additionally showed:

• an empty markdown cell ({"type":"markdown","text":""}) after every timeseries cell, each with a fresh id per GET; updates whose payload omits these cells do not durably remove them;
• the same cell id appearing twice within a single GET response (4 duplicated ids among 12 timeseries cells).

A freshly created notebook does not show either symptom (single, repeated, rapid-succession, id-bearing and id-stripped updates all round-trip cleanly), so these look like a server-side state/merge condition. Reported together because the markdown id instability above makes id-based diagnosis or repair of such a notebook impossible via the API.

Environment

• OS: macOS 26 (aarch64)
• Pup version: 0.65.4 (rust 1.95.0)
• Auth: OAuth2, site datadoghq.com
• Reproduces identically via pup api v1/notebooks/{id} raw passthrough (server-side, cf. [BUG] pup notebooks update doubles every backslash in markdown cells on each round trip #554)