Tests for get_dd_changelog and signal_analytics, plus two CI observations

[721AndrewWang]

Hi Simon,

I have been reading imas-codex over the last few days (the fork at
Simon-McIntosh/imas-codex, since that is where the current code lives) and
would like to contribute a small, test-only change. Before opening a PR I
want to check scope and a couple of behaviours with you.

What I have done

Going through the tools registered in imas_codex/llm/server.py and
grepping tests/ for each tool name, three had no test references:
get_dd_changelog, get_dd_migration_guide (its migration_guide module
is covered, only the handler is not) and signal_analytics.

I wrote mock-backed unit tests for the first and the last, no source
changes:

• tests/tools/test_dd_changelog.py (15 tests): VersionTool.get_dd_changelog
  parameter contract, result shaping, error path; format_dd_changelog_report
  header, version-range labels, rows, truncation hint.
• tests/llm/test_signal_analytics.py (33 tests): group_by validation,
  routing between the plain query and the CHECKED_WITH join, check_status
  and error_type filters, parameterisation, error handling, _format_analytics.

Branch: https://github.com/721AndrewWang/IMAS-Codex/tree/test/dd-changelog-signal-analytics
(based on your fork's main at 7745119). Locally: 48 passed, ruff check
and ruff format --check clean, tests/tools + tests/llm still 496 passed.

Two behaviours I pinned but did not change

1. _signal_analytics rejects unknown group_by values but silently drops
   unknown filters keys. The test asserts the current behaviour and says so
   in its docstring. If you would rather it error like group_by, I can
   change the tool and flip the test in the same PR.
2. get_dd_changelog compares versions as strings in Cypher
   (change.version > $from_version) while the same file has
   _version_sort_key for numeric ordering. It works for the DD versions that
   exist today (3.22 to 3.42, 4.x); flagging it in case it is unintended.

Two CI observations on the fork

• The last 40 Test runs all fail at "Format check with ruff". With the
  locked ruff==0.15.1, two files need formatting:
  tests/standard_names/test_detachment_attribution_wiring.py and
  tests/standard_names/test_sourceless_name_recovery.py. Happy to include
  that in the PR or leave it to you.
• Because those runs stop at the format step, the pytest step has not run
  recently. Locally, the CI command uv sync --extra test --no-dev produces
  an environment where tests/conftest.py fails to import
  (ModuleNotFoundError: No module named 'imas_standard_names'), since
  imas-standard-names is only declared in the dev group. Syncing with the
  dev group works. This may bite once the format step is green.

Questions

• Should the PR target this repository or Simon-McIntosh/imas-codex?
• Anything in the test layout you would like done differently?

Thanks,
Jianan Wang