feat(opt): three efficiency levers for NL→SQL workflow

Lever 1: Join-graph via FK logics in annotations + suggest-fks command
  - TableAnnotation.foreign_keys field stores curated logical FKs (DWH has no FK constraints)
  - tht schema suggest-fks: mine from approved SQL, heuristics (time_key → dim_time),
    same-name discovery + explicit --assume flag for multi-owner PKs
  - mschema renders 【Foreign keys】 section populated; validation in merge.py
  - SKILL.md F4 now reads FKs from mschema-text, no custom data_time_key logic

Lever 2: Context-pack consolidation at kickoff (tht search pack)
  - Single embedding of question, reused for schema + evidence + solved searches
  - One command: tht search pack <question> --session <id> → retrieval_pack.md
  - Graceful degradation when Ollama/vector store unreachable (exit 0, empty sections)
  - SKILL.md F1 prescribes as first call; reduces model thinking turns via pre-retrieval

Lever 3: Phase-summary recap v2 auto-construction from session ledger
  - tht session show --json includes full decisions ledger
  - tht phase meta --json exports 'emits' (substantive decision types per phase)
  - Gate appends deterministic 【Decisioni registrate in questa fase】 section (appendLedgerSection)
  - Model authors only summary + checks; recap table comes from persisted state (exact by construction)
  - SKILL.md Disciplina 6: brief model output, gate fills the rest

Tests: 358 Python (including 10 FK + 3 pack + 1 session-ledger tests) + 111 JS gate tests, all pass.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

harness/tests/test_search_pack.py

@@ -0,0 +1,115 @@
+import json
+from datetime import datetime
+from types import SimpleNamespace
+
+from typer.testing import CliRunner
+
+from tht.cli import app
+from tht.mschema.models import ColumnPhysical, PhysicalSchema, TablePhysical
+from tht.vectorstore.embeddings import EmbeddingsError
+
+
+class _FakeEmbedder:
+    def __init__(self):
+        self.calls = 0
+
+    def embed_query(self, text):
+        self.calls += 1
+        return [0.1, 0.2, 0.3]
+
+
+class _FakeSearcher:
+    def search(self, vec, top_n, kinds=None):
+        if kinds == ["solved_question"]:
+            return [SimpleNamespace(
+                kind="memory", ref="s-1", id="m1", title="q solved",
+                similarity=0.91, content="quanti pazienti nel 2024?",
+                metadata={"session_id": "2026-01-01-000000-x", "sql": "SELECT 1",
+                          "tables": ["fact_ablazione"], "question": "quanti pazienti nel 2024?"},
+            )]
+        if kinds == ["schema_table", "schema_column"]:
+            return [SimpleNamespace(
+                kind="schema_table", ref="fact_ablazione", id="t1",
+                title="Tabella fact_ablazione", similarity=0.88,
+                content="Tabella fact_ablazione", metadata={},
+            )]
+        if kinds == ["evidence"]:
+            return [SimpleNamespace(
+                kind="evidence", ref="ev1", id="ev1", title="Dominio ablazione",
+                similarity=0.8, content="L'ablazione e' una procedura...",
+                metadata={"status": "approved"},
+            )]
+        return []
+
+
+def _workspace(tmp_path, with_session=None):
+    PhysicalSchema(
+        database="d", schema="s", introspected_at=datetime(2026, 1, 1),
+        tables={"fact_ablazione": TablePhysical(
+            comment="Ablazioni", columns={"cod_paz": ColumnPhysical(type="bigint")})},
+    ).to_yaml(tmp_path / "artifacts" / "mschema" / "physical.yaml")
+    cfg = tmp_path / "workspace.yaml"
+    cfg.write_text(
+        "database: {database: d, schema: s, user: u, password: p, transport: direct}\n"
+        "vector_db: {database: v, schema: public, user: u, password: p}\n"
+        "embeddings: {base_url: 'http://localhost:11434', model: nomic-embed-text, dim: 8}\n"
+        f"paths: {{artifacts: {tmp_path/'artifacts'}, indexes: {tmp_path/'i'}, "
+        f"sessions: {tmp_path/'sessions'}}}\n"
+    )
+    if with_session:
+        sdir = tmp_path / "sessions" / with_session
+        sdir.mkdir(parents=True)
+        (sdir / "session_manifest.yaml").write_text(
+            f"id: {with_session}\nquestion: q\ndatabase: d\nschema: s\n"
+            "created_at: 2026-01-01T00:00:00+00:00\nstatus: open\n"
+        )
+    return cfg
+
+
+def _patch(monkeypatch, embedder, searcher):
+    import tht.cli.vector_cmd as vc
+
+    monkeypatch.setattr(vc, "make_embedder", lambda _cfg: embedder)
+    monkeypatch.setattr(vc, "open_searcher", lambda _cfg: searcher)
+
+
+def test_pack_single_embed_and_sections(tmp_path, monkeypatch):
+    cfg = _workspace(tmp_path)
+    emb = _FakeEmbedder()
+    _patch(monkeypatch, emb, _FakeSearcher())
+    res = CliRunner().invoke(app, ["search", "pack", "quanti pazienti", "-c", str(cfg)])
+    assert res.exit_code == 0, res.output
+    assert emb.calls == 1  # UN solo embedding per le tre ricerche
+    assert "fact_ablazione" in res.output and "Ablazioni" in res.output
+    assert "Dominio ablazione" in res.output
+    assert "SELECT 1" in res.output
+
+
+def test_pack_json_and_session_file(tmp_path, monkeypatch):
+    sid = "2026-01-01-000000-test"
+    cfg = _workspace(tmp_path, with_session=sid)
+    _patch(monkeypatch, _FakeEmbedder(), _FakeSearcher())
+    res = CliRunner().invoke(
+        app, ["search", "pack", "q", "-c", str(cfg), "--session", sid, "--json"]
+    )
+    assert res.exit_code == 0, res.output
+    data = json.loads(res.output)
+    assert data["tables"][0]["name"] == "fact_ablazione"
+    pack = tmp_path / "sessions" / sid / "retrieval_pack.md"
+    assert pack.exists()
+    assert "Retrieval pack" in pack.read_text()
+
+
+def test_pack_degrades_gracefully(tmp_path, monkeypatch):
+    cfg = _workspace(tmp_path)
+
+    class _Broken:
+        def embed_query(self, text):
+            raise EmbeddingsError("ollama down")
+
+    _patch(monkeypatch, _Broken(), _FakeSearcher())
+    res = CliRunner().invoke(app, ["search", "pack", "q", "-c", str(cfg), "--json"])
+    assert res.exit_code == 0, res.output
+    data = json.loads(res.output[res.output.index("{"):])
+    assert data["tables"] == [] and data["evidence"] == [] and data["solved"] == []
+    assert any("retrieval non disponibile" in w for w in data["warnings"])