refactor(ai): remove semantic embedding cache + Qdrant vector store

Hapus seluruh fitur embedding/Qdrant (tidak dipakai lagi):

- gateway: drop embeddingClient.ts, qdrantClient.ts, archiveEmbedder.ts
  dan tes qdrantEnsure.test.ts; moderationOrchestrator kembali ke
  exact-hash cache -> LLM (tanpa phase-2 semantic lookup); textCacheStore
  kehilangan findSimilarTextModeration / parseQdrantVerdict /
  isSemanticBandAccepted / upsertBareKeyToQdrant; cache-prune hanya
  menyapu Postgres.
- backend: drop embed.ts + qdrant.ts, endpoint messages.semanticSearch
  dan schema/type terkait; kolom embedding dilepas dari schema
  text_analysis_cache.
- frontend: hapus toggle EXACT/SEMANTIC, hook useSemanticSearch,
  API client + tipe SemanticSearchResult.
- config: buang AI_LLM_EMBEDDING_* dan QDRANT_* (env + .env.example).
- docs: ARCHITECTURE.md / AGENTS.md / README.md / diagram arsitektur
  disesuaikan (LLM caller - vision, cache = exact-hash saja).

Verifikasi: tsc 0 (backend, gateway, frontend); bun test 135 pass +
37 pass, 0 fail; biome 0 error.

services/discord-gateway/tests/cacheGuards.test.ts

@@ -1,11 +1,9 @@
 // ═══════════════════════════════════════════════════════════════════════════
-// Semantic two-band acceptance + global exact-cache reuse guard
+// Global exact-cache reuse guard
 // ═══════════════════════════════════════════════════════════════════════════
 // Design (2026-08-24): cache hits may be served MORE aggressively for
 // verdicts that cannot trigger enforcement actions, and NEVER more
-// aggressively for actionable ones. Two layers enforce this:
-//   - isSemanticBandAccepted: similarity thresholds differ by verdict class
-//     (clean band 0.92 default vs strict actionable band 0.97 default).
+// aggressively for actionable ones:
 //   - isGloballyReusableCleanVerdict: context-free (cross-channel) reuse of
 //     the legacy bare key only for clean / flagless / action=none verdicts
 //     with high confidence and bounded age.
@@ -31,64 +29,6 @@ function makeVerdict(
   };
 }
 
-describe("isSemanticBandAccepted", () => {
-  it("accepts a non-actionable clean verdict at the loose clean band", () => {
-    // Default AI_LLM_EMBEDDING_MIN_SIMILARITY_CLEAN = 0.92.
-    expect(isBandAccept(makeVerdict(), 0.93)).toBe(true);
-  });
-
-  it("accepts a clean verdict exactly at the clean band boundary", () => {
-    expect(isBandAccept(makeVerdict({ confidence: 0.99 }), 0.92)).toBe(true);
-  });
-
-  it("rejects a clean verdict below the clean band", () => {
-    expect(isBandAccept(makeVerdict(), 0.91)).toBe(false);
-  });
-
-  it("rejects an actionable flagged verdict between the bands", () => {
-    // 0.93 >= clean band BUT < strict band → must NOT be served.
-    expect(
-      isBandAccept(
-        makeVerdict({ status: "flagged", flags: ["hate_speech"] }),
-        0.93,
-      ),
-    ).toBe(false);
-  });
-
-  it("accepts a flagged verdict at the strict band", () => {
-    expect(
-      isBandAccept(
-        makeVerdict({ status: "flagged", flags: ["hate_speech"] }),
-        0.98,
-      ),
-    ).toBe(true);
-  });
-
-  it("rejects a warn verdict below the strict band", () => {
-    expect(
-      isBandAccept(
-        makeVerdict({ status: "warn", recommendedAction: "warn" }),
-        0.96,
-      ),
-    ).toBe(false);
-  });
-
-  it("treats a clean verdict WITH flags as actionable (strict band)", () => {
-    expect(isBandAccept(makeVerdict({ flags: ["borderline"] }), 0.93)).toBe(
-      false,
-    );
-  });
-
-  it("treats a clean verdict with a non-none action as actionable", () => {
-    expect(
-      isBandAccept(makeVerdict({ recommendedAction: "review" }), 0.93),
-    ).toBe(false);
-  });
-});
-
-// Import indirection so the describe block reads cleanly.
-import { isSemanticBandAccepted as isBandAccept } from "../src/modules/ai-moderation/textCacheStore.js";
-
 describe("isGloballyReusableCleanVerdict", () => {
   it("accepts a fresh, confident, flagless clean verdict", () => {
     const v = makeVerdict({ confidence: 0.9 });

services/discord-gateway/tests/qdrantEnsure.test.ts

@@ -1,82 +0,0 @@
-import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
-
-// The qdrant client reads config at import time — we only need the
-// reset-on-failure behavior, so mock global fetch and import fresh.
-import {
-  ensureQdrantCollection,
-  resetQdrantCollectionCache,
-} from "../src/modules/ai-moderation/qdrantClient.js";
-
-function jsonResponse(body: unknown, ok = true, status = 200): Response {
-  return {
-    ok,
-    status,
-    text: async () => JSON.stringify(body),
-  } as unknown as Response;
-}
-
-describe("ensureQdrantCollection retry-on-failure", () => {
-  beforeEach(() => {
-    resetQdrantCollectionCache();
-    vi.restoreAllMocks();
-  });
-
-  afterEach(() => {
-    resetQdrantCollectionCache();
-    vi.restoreAllMocks();
-  });
-
-  it("returns false on failure and does NOT get stuck — retries on next call", async () => {
-    // First call: GET collection fails hard (not a 404) → ensure rejects.
-    const fetchMock = vi
-      .spyOn(globalThis, "fetch")
-      .mockRejectedValueOnce(new Error("network down"))
-      // Second call: GET returns a matching-size collection → success.
-      .mockResolvedValueOnce(
-        jsonResponse({
-          result: { config: { params: { vectors: { size: 3072 } } } },
-        }),
-      );
-
-    const first = await ensureQdrantCollection(3072);
-    expect(first).toBe(false);
-
-    // Without the fix, the second call returns the memoised rejected promise
-    // and fetch is never called again. With the fix, it retries.
-    const second = await ensureQdrantCollection(3072);
-    expect(second).toBe(true);
-    expect(fetchMock).toHaveBeenCalledTimes(2);
-  });
-
-  it("recreates collection when vector size changes (DELETE + PUT)", async () => {
-    const fetchMock = vi
-      .spyOn(globalThis, "fetch")
-      .mockResolvedValueOnce(
-        jsonResponse({
-          result: { config: { params: { vectors: { size: 2048 } } } },
-        }),
-      ) // GET old size
-      .mockResolvedValueOnce(jsonResponse({ result: true })) // DELETE
-      .mockResolvedValueOnce(jsonResponse({ result: true })); // PUT
-
-    const ok = await ensureQdrantCollection(3072);
-    expect(ok).toBe(true);
-
-    const methods = fetchMock.mock.calls.map(
-      (c) => (c[1] as RequestInit).method,
-    );
-    expect(methods).toEqual(["GET", "DELETE", "PUT"]);
-  });
-
-  it("is idempotent when the collection already matches", async () => {
-    const fetchMock = vi.spyOn(globalThis, "fetch").mockResolvedValueOnce(
-      jsonResponse({
-        result: { config: { params: { vectors: { size: 3072 } } } },
-      }),
-    );
-
-    const ok = await ensureQdrantCollection(3072);
-    expect(ok).toBe(true);
-    expect(fetchMock).toHaveBeenCalledTimes(1);
-  });
-});

services/discord-gateway/tests/storedStatusNormalization.test.ts

@@ -6,10 +6,10 @@
 // ["conflict_instigation"]) fell into the legacy `flags.length === 0 ?
 // clean : flagged` branch and was read back as FLAGGED. Downstream this
 // broke auto-delete eligibility gating and mislabelled warnings on the
-// dashboard. parseQdrantVerdict had the same narrowing (warn → clean).
+// dashboard.
 //
-// Fix: normalizeStoredStatus() accepts the full clean/warn/flagged union in
-// BOTH readers; unknown/legacy values still derive from flags.
+// Fix: normalizeStoredStatus() accepts the full clean/warn/flagged union;
+// unknown/legacy values still derive from flags.
 import { describe, expect, it } from "vitest";
 import { normalizeStoredStatus } from "../src/modules/ai-moderation/textCacheStore.js";