refactor(mcp): 16 → 5 tool aggregati (dataset, query, pipeline, source, contract)

[Gabrymi93]

Sostituisce i 16 tool flat con 5 tool CLI-like

Ogni tool ha un parametro action che fa da dispatch, riducendo la complessita' per gli agenti AI da 16 nomi da ricordare a 5 pattern uniformi.

Nuovi tool

Tool	Azioni	Params chiave
toolkit_dataset	find, overview, status, preflight, schema-diff	action + query/slug/config_path
toolkit_query	run, preview	action + datasets/sql o url
toolkit_pipeline	contract, runs, registry_list, registry_show, graph	action + layer/config_path/repo
toolkit_source	probe, ckan, links, sparql	action + url/endpoint/query
toolkit_contract	(unico)	layer (backward compat)

Esempi

# Prima (3 tool diversi)
toolkit_find(query="terna")
toolkit_dataset_overview(slug="terna_electricity_by_source")
toolkit_layer(datasets=["terna_electricity_by_source"], sql="SELECT ...")

# Dopo (2 tool, pattern uniforme)
toolkit_dataset(action="find", query="terna")
toolkit_dataset(action="overview", slug="terna_electricity_by_source")
toolkit_query(action="run", datasets=["terna_electricity_by_source"], sql="SELECT ...")

Metriche

• Test: 34/34 passano
• Ruff: clean
• File changed: 2 (server.py, test_mcp_server.py)
• LOC: +505 / -477

Note

• toolkit_contract mantenuto per backward compat
• Le implementazioni *_ops.py non sono cambiate — solo il layer MCP in server.py
• Default mode="sql" per toolkit_query (catalog mode lo richiede)