feat: Phase 2 — embeddings + hybrid memory.search

Adds a small embedder sidecar (Xenova/bge-small-en-v1.5, ONNX, CPU-only)
that the web app calls inline on memory.write and memory.update, and on
demand from the new memory.search tool.

memory.search performs three candidate fetches in parallel — pgvector
cosine similarity, Postgres full-text via plainto_tsquery + ts_rank_cd,
and tag-set overlap — then fuses them with Reciprocal Rank Fusion
(k=60). Each result carries its per-source rank so the model can see
*why* a memory surfaced.

The migrator boot step gained an idempotent embedding backfill: any row
with embedding IS NULL is batched (32 at a time) through the embedder
after SQL migrations apply. Safe to run on every boot.

New tool memory.update fixes the missing edit path; centralises the
re-embed-on-content-change rule alongside write.

Stack additions:
- apps/embedder/ — Fastify server, persistent /data/models volume so the
  ~30 MB model only downloads once
- apps/web/lib/embedder.ts — typed HTTP client with batched embed +
  health probe
- packages/schemas — MemoryUpdateInput, MemorySearchInput
- docker-compose — embedder service, healthcheck, app + migrator both
  depend_on it healthy; EMBEDDER_URL promoted to a required env var

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/web/lib/embedder.ts

@@ -0,0 +1,65 @@
+import { env } from "@/lib/env";
+
+/**
+ * Thin HTTP client for the embedder sidecar. Used by memory.write /
+ * memory.update / memory.search and by the migrator's backfill step.
+ *
+ * Calls are blocking on purpose — write-path latency is a worthwhile
+ * trade for "the memory I just wrote is searchable now."
+ */
+
+export class EmbedderError extends Error {
+  constructor(message: string, public readonly status?: number) {
+    super(message);
+    this.name = "EmbedderError";
+  }
+}
+
+function url(): string {
+  const u = env().EMBEDDER_URL;
+  if (!u) throw new EmbedderError("EMBEDDER_URL is not configured");
+  return u.replace(/\/$/, "");
+}
+
+/** Embed a batch of texts. Returns one vector per input. */
+export async function embedTexts(texts: string[]): Promise<number[][]> {
+  if (texts.length === 0) return [];
+
+  const res = await fetch(`${url()}/embed`, {
+    method: "POST",
+    headers: { "content-type": "application/json" },
+    body: JSON.stringify({ texts }),
+  });
+
+  if (!res.ok) {
+    const detail = await res.text().catch(() => "");
+    throw new EmbedderError(
+      `embedder returned ${res.status}: ${detail.slice(0, 200)}`,
+      res.status,
+    );
+  }
+  const body = (await res.json()) as { vectors: number[][] };
+  if (!Array.isArray(body.vectors) || body.vectors.length !== texts.length) {
+    throw new EmbedderError("embedder response shape mismatch");
+  }
+  return body.vectors;
+}
+
+/** Embed a single text — convenience for one-off calls. */
+export async function embedText(text: string): Promise<number[]> {
+  const [vec] = await embedTexts([text]);
+  if (!vec) throw new EmbedderError("embedder returned no vector");
+  return vec;
+}
+
+/** Quick check used by the migrator before backfilling. */
+export async function embedderReady(): Promise<boolean> {
+  try {
+    const res = await fetch(`${url()}/health`);
+    if (!res.ok) return false;
+    const body = (await res.json()) as { ready?: boolean };
+    return body.ready === true;
+  } catch {
+    return false;
+  }
+}