`describegpt`: make it CKAN-aware

[jqnatividad]

So users can customize LLM prompts to refer to associated CKAN instances:

• to fetch the Dataset from CKAN if it's not available locally
• to use existing Data Dictionaries, Summary Stats and Frequency tables (in case they have DP+ with the DRUF installed)
• to issue catalog searches
• to fetch for additional metadata context (both DCAT 3 and Croissant)
• to check for related datasets (e.g. while looking up an NYC address; check NYPD COMPstat data)
• to check and use controlled vocabularies
• to check for existing tags, themes and groups
• to get lookup tables, and enrich/normalize LLM answers from these tables (e.g. with the Building ID number, get the address; with an NYC address, get associated data (e.g. Community Board, Borough, Police and Fire Precincts, Congressional District, Census tables, etc.) )
• to check private datasets, if provided with the requisite CKAN token

Preferably, via MCP and/or tool-use.

[jqnatividad]

Also, if a --prompt cannot be answered with the existing Dataset. Use its CKAN connection to suggest Datasets that may be able to answer the prompt.

see #2944

[jqnatividad]

If a --prompt provides a CKAN URL instead of a local Dataset, search the CKAN Catalog and auto-select the Dataset to answer the prompt based on its DCAT3/Croissant metadata.