feat(llm): add embedding adapter with cache and similarity utils (S1.3)

Add OpenAI-compatible embedding support (works with both OpenAI and
OpenRouter), file-based embedding cache with content-digest invalidation,
and pure-Python cosine similarity utilities for downstream redundancy
detection.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

markitect/llm/embedding_cache.py

@@ -0,0 +1,64 @@
+"""
+File-based embedding cache.
+
+Stores embedding vectors in a single JSON file keyed by entity slug.
+Each entry includes a content digest so stale embeddings are
+automatically invalidated when entity content changes.
+"""
+
+import json
+from pathlib import Path
+from typing import Optional
+
+
+class EmbeddingCache:
+    """Persistent cache for embedding vectors.
+
+    Structure on disk (``embeddings.json``)::
+
+        {
+            "division-of-labour": {"digest": "abc123", "vector": [0.1, ...]},
+            ...
+        }
+    """
+
+    def __init__(self, cache_dir: Path):
+        self._path = cache_dir / "embeddings.json"
+        self._data: dict[str, dict] = {}
+        self._hits = 0
+        self._misses = 0
+        self._load()
+
+    def get(self, slug: str, content_digest: str) -> Optional[list[float]]:
+        """Return the cached vector if *content_digest* matches, else ``None``."""
+        entry = self._data.get(slug)
+        if entry is not None and entry.get("digest") == content_digest:
+            self._hits += 1
+            return entry["vector"]
+        self._misses += 1
+        return None
+
+    def put(self, slug: str, content_digest: str, vector: list[float]) -> None:
+        """Store or overwrite the embedding for *slug*."""
+        self._data[slug] = {"digest": content_digest, "vector": vector}
+
+    def save(self) -> None:
+        """Write cache to disk."""
+        self._path.parent.mkdir(parents=True, exist_ok=True)
+        self._path.write_text(json.dumps(self._data, separators=(",", ":")))
+
+    def stats(self) -> dict:
+        """Return cache statistics."""
+        return {
+            "entries": len(self._data),
+            "hits": self._hits,
+            "misses": self._misses,
+        }
+
+    def _load(self) -> None:
+        """Read cache from disk if it exists."""
+        if self._path.is_file():
+            try:
+                self._data = json.loads(self._path.read_text())
+            except (json.JSONDecodeError, OSError):
+                self._data = {}