feat(harness): port vectorstore dual-key + reader RPC (D11, §5.4)

Ports vectorstore/{rest_client,rest_writer,store,reader,embeddings,records},
evidence/model (leaf dep of records), and cli/_guards (require_vector_write_allowed
workstation write-guard). Renamed psdwp3->nsp, verbatim.

VectorRestClient gains an api_key property so reader/writer clients carry their
distinct keys visibly (spec D11: vector_reader / vector_writer on the same endpoint).

scripts/create_vector_reader_rpc.sql is NEW: the reader RPCs (search_similar,
list_tables) lived server-side in Supabase and were never versioned. Authored now
mirroring the writer allowlist pattern (table allowlist, security definer, revoke
from anon/authenticated, grant to vector_reader only). Writer RPC ported verbatim.

L1: test_vector_dual_key (7 tests) pins the dual-key construction + the workstation
write-guard (exit 4 without writer key).

harness/nsp/evidence/model.py

@@ -0,0 +1,62 @@
+from pathlib import Path
+from typing import Literal
+
+import yaml
+from pydantic import BaseModel, ValidationError
+
+
+class EvidenceError(Exception):
+    pass
+
+
+class EvidenceDoc(BaseModel):
+    id: str
+    title: str
+    # tier e status sono opzionali: la sola presenza di un documento basta a
+    # vettorizzarlo, quindi l'autore ETL non e' obbligato a compilarli.
+    tier: Literal["structural", "concept"] = "structural"
+    status: Literal["auto", "draft", "reviewed"] = "reviewed"
+    sources: list[str] = []
+    tables: list[str] = []
+    concepts: list[str] = []
+    body: str = ""
+    path: Path | None = None  # valorizzato al load, escluso dal dump
+
+    @classmethod
+    def parse(cls, text: str, path: Path | None = None) -> "EvidenceDoc":
+        if not text.startswith("---\n"):
+            raise EvidenceError(f"frontmatter mancante in {path or '<testo>'}")
+        try:
+            _, fm, body = text.split("---\n", 2)
+        except ValueError as e:
+            raise EvidenceError(f"frontmatter malformato in {path or '<testo>'}") from e
+        meta = yaml.safe_load(fm)
+        if not isinstance(meta, dict):
+            raise EvidenceError(f"frontmatter non valido in {path or '<testo>'}")
+        try:
+            return cls.model_validate({**meta, "body": body.strip("\n"), "path": path})
+        except ValidationError as e:
+            raise EvidenceError(f"evidence non valida in {path or '<testo>'}:\n{e}") from e
+
+    def dump(self) -> str:
+        meta = self.model_dump(exclude={"body", "path"}, mode="json")
+        fm = yaml.safe_dump(meta, sort_keys=False, allow_unicode=True)
+        return f"---\n{fm}---\n{self.body}\n"
+
+    def save(self, path: Path) -> None:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.write_text(self.dump())
+        self.path = path
+
+
+def load_evidence_dir(root: Path) -> list[EvidenceDoc]:
+    """Carica ricorsivamente tutte le evidence sotto `root`, preservando la
+    gerarchia per dominio. I file README (di sola navigazione) sono ignorati."""
+    docs: list[EvidenceDoc] = []
+    if not root.is_dir():
+        return docs
+    for f in sorted(root.rglob("*.md")):
+        if f.name.upper().startswith("README"):
+            continue
+        docs.append(EvidenceDoc.parse(f.read_text(), path=f))
+    return docs