test(8.0): Tier-1 integration via deterministic embedder (suite runnable again)

The integration suite loaded the real WASM model for every test, so a full run
took ~hours and reliably aborted (vitest reporter timeout) — which is why it was
never gated and silently rotted. Brainy already separates the embedder from all
other logic, so running integration with a deterministic embedder keeps every
code path real (HNSW build+search, graph traversal, metadata filtering,
transactions, storage/sharding, VFS, aggregation, import) while making the suite
complete in ~2 min on ANY machine — no GPU, no 16 GB requirement.

- Add isDeterministicEmbedMode() — a single source of truth for the test-embedder
  switch (BRAINY_DETERMINISTIC_EMBEDDINGS, plus the legacy BRAINY_UNIT_TEST alias),
  used by EmbeddingManager (init/embed/embedBatch) and brainy's eager-init guard.
- setup-integration.ts opts into it: Tier 1 = functional end-to-end. Real-model
  semantic quality + the real embedding pipeline move to a Tier-2 suite.

Full integration now runs 482 passing / 583 in ~2 min (was un-completable). The
remaining failures are pre-existing test rot, fixed next. Unit 1414 green.

tests/setup-integration.ts

@@ -1,60 +1,44 @@
 /**
- * Integration Test Setup - REAL AI functionality
- * 
- * This setup enables real AI models for integration testing
- * Requires high memory environment (16GB+ RAM)
+ * Integration Test Setup — Tier 1: Functional end-to-end (deterministic embedder)
+ *
+ * These tests exercise every real code path — HNSW build + search, graph
+ * traversal, metadata filtering, triple-intelligence composition, MVCC
+ * transactions, sharding/persistence, VFS, aggregation, import — but with a
+ * DETERMINISTIC embedder instead of the real WASM transformer. The embedding's
+ * semantic *quality* is irrelevant to these assertions (which are structural:
+ * counts, ids, relationships, self-retrieval, transaction outcomes); making the
+ * vector deterministic turns a multi-hour, un-completable suite into a sub-minute
+ * one that any contributor can run on any machine, no GPU or 16 GB required.
+ *
+ * Semantic relevance + the real embedding pipeline are covered separately by the
+ * Tier-2 semantic suite (`tests/semantic/`, real model) — see
+ * `tests/configs/vitest.semantic.config.ts`.
  */
 
-beforeAll(async () => {
-  console.log('🤖 Integration Test Environment: Using REAL AI models')
-  console.log('⚠️  Requires 16GB+ RAM - this is normal for AI testing')
-  
-  // Set up environment for real AI testing
-  process.env.BRAINY_INTEGRATION_TEST = 'true'
-  process.env.BRAINY_MODELS_PATH = './models'
-  process.env.BRAINY_ALLOW_REMOTE_MODELS = 'false'  // Use local models only
-  
-  // Set memory limits and optimizations
-  process.env.ORT_DISABLE_MEMORY_ARENA = '1'
-  process.env.ORT_DISABLE_MEMORY_PATTERN = '1'
-  process.env.ORT_INTRA_OP_NUM_THREADS = '2'
-  process.env.ORT_INTER_OP_NUM_THREADS = '2'
-  
-  // Mark as integration test environment
-  ;(globalThis as any).__BRAINY_INTEGRATION_TEST__ = true
-  
-  // Check memory availability
-  const availableMemoryGB = process.env.NODE_OPTIONS?.includes('max-old-space-size')
-    ? parseInt(process.env.NODE_OPTIONS.match(/--max-old-space-size=(\d+)/)?.[1] || '0') / 1024
-    : 4
-  
-  console.log(`📊 Node.js heap limit: ${availableMemoryGB.toFixed(1)}GB`)
-  
-  if (availableMemoryGB < 8) {
-    console.warn('⚠️  WARNING: Less than 8GB allocated for integration tests')
-    console.warn('   Recommended: NODE_OPTIONS="--max-old-space-size=16384"')
-    console.warn('   Tests may fail due to insufficient memory')
-  }
-}, 60000) // 1 minute timeout for setup
+beforeAll(() => {
+  console.log('🧪 Integration (Tier 1): deterministic embedder, all other code paths real')
 
-afterAll(async () => {
-  // Clean up
-  delete process.env.BRAINY_INTEGRATION_TEST
-  delete (globalThis as any).__BRAINY_INTEGRATION_TEST__
-  
-  // Force garbage collection if available
-  if (global.gc) {
-    global.gc()
-  }
-}, 30000) // 30 second timeout for cleanup
+  // Deterministic embeddings — see src/embeddings/deterministicEmbedMode.ts.
+  process.env.BRAINY_DETERMINISTIC_EMBEDDINGS = 'true'
+  ;(globalThis as { __BRAINY_DETERMINISTIC_EMBED__?: boolean }).__BRAINY_DETERMINISTIC_EMBED__ = true
 
-// Utility function to skip tests if not enough memory
-export function requiresMemory(minGB: number) {
-  const availableMemoryGB = process.env.NODE_OPTIONS?.includes('max-old-space-size')
-    ? parseInt(process.env.NODE_OPTIONS.match(/--max-old-space-size=(\d+)/)?.[1] || '0') / 1024
-    : 4
-  
-  if (availableMemoryGB < minGB) {
-    throw new Error(`Test requires ${minGB}GB memory, only ${availableMemoryGB.toFixed(1)}GB allocated`)
-  }
-}
+  // Marker retained for any test that branches on the integration environment.
+  ;(globalThis as { __BRAINY_INTEGRATION_TEST__?: boolean }).__BRAINY_INTEGRATION_TEST__ = true
+})
+
+afterAll(() => {
+  delete process.env.BRAINY_DETERMINISTIC_EMBEDDINGS
+  delete (globalThis as { __BRAINY_DETERMINISTIC_EMBED__?: boolean }).__BRAINY_DETERMINISTIC_EMBED__
+  delete (globalThis as { __BRAINY_INTEGRATION_TEST__?: boolean }).__BRAINY_INTEGRATION_TEST__
+})
+
+/**
+ * No-op retained for back-compat. Tier-1 integration tests use the deterministic
+ * embedder, so the real model's 16 GB requirement no longer applies — there is
+ * nothing to gate on. (Real-model memory cost lives in the Tier-2 semantic suite.)
+ *
+ * @param _minGB - Ignored.
+ */
+export function requiresMemory(_minGB: number): void {
+  // Intentionally empty — see the doc comment above.
+}