Centralize sample-dataset metadata (load_sample / tests / docs / MANIFEST.in)

[SaguaroDev]

Description

In #771 (review of #745), @henrydingliu observed that sample-dataset column names are duplicated across three places:

• the load_sample branch in chainladder/utils/utility_functions.py
• the test that exercises the load (e.g. test_load_sample_clrd2025)
• the dataset table in docs/library/sample_data.md

He suggested storing the metadata of available sample datasets and fields in one place, and using it to:

• run load_sample()
• key off tests
• generate MANIFEST.in
• generate sample_data.md

This applies to every sample dataset in chainladder/utils/data/, not just clrd2025 — the same triple-duplication exists today for clrd, berqsherm, xyz, the friedland family, etc. The fix is general:

1. Define one manifest file (YAML or a Python dict in chainladder/utils/data/_manifest.py) keyed by sample name with origin, development, index, columns, cumulative, and any per-sample flags.
2. Refactor load_sample to look up its config from the manifest rather than the long if key.lower() == ... chain.
3. Generate the sample_data.md table at docs-build time (or commit the generated file with a regen script under scripts/).
4. Have MANIFEST.in include chainladder/utils/data/*.csv via the manifest's listed files, or just keep the existing wildcard — whichever the maintainers prefer.
5. Update tests to iterate over the manifest, so adding a new sample is a one-line change.

Is your feature request aligned with the scope of the package?

• Yes, absolutely!

Describe the solution you'd like, or your current workaround.

See above. Current workaround is the existing pattern — every new sample dataset (including #745's clrd2025) updates three files by hand.

Do you have any additional supporting notes?

Filed at @henrydingliu's suggestion in #771 (comment). Keeping #771 / #745 scoped to landing the data; this refactor is its own piece of work.