feat: TS SDK v3 port, graph store removal, client v3 API migration

TypeScript SDK v3 pipeline (full parity with Python):
- Single-pass additive extraction with ADDITIVE_EXTRACTION_PROMPT
- Hybrid search (semantic + BM25 + entity boost) with additive scoring
- 8-phase batch pipeline (batch embed, persist, entity linking)
- New utils: scoring.ts, lemmatization.ts (natural), entity_extraction.ts (compromise)
- keywordSearch() on 8 vector stores (3 full: PGVector, Memory, Azure AI Search)
- Message persistence in SQLiteManager (rolling window of 10)
- Entity store as second vector collection
- MAX_BATCH=100 chunking guard on OpenAI/Azure embedBatch
- Updated default LLM model to gpt-4.1-nano-2025-04-14
- compromise + natural added as peer dependencies

Graph store removal (Python + TypeScript):
- Removed Neo4j, Memgraph, Kuzu, Neptune, Apache AGE integrations
- Deleted 18 graph-related files across both SDKs
- Removed GraphStoreFactory, GraphStoreConfig, graph_store config field
- Removed "relations" key from all API responses
- Removed graph optional dependency group from pyproject.toml
- Removed neo4j-driver from TS peerDependencies
- Simplified add/search/delete/reset (no more parallel graph operations)

Client SDK v3 API migration:
- add() endpoint: /v1/memories/ -> /v3/memories/ (async response)
- search() endpoint: /v2/memories/search/ -> /v3/memories/search/
- Removed output_format injection and v1.1 unwrapping logic
- Applied to both Python (sync + async) and TypeScript clients

Review feedback fixes:
- Removed deprecated custom_update_memory_prompt from MemoryConfig
- Added MAX_BATCH=100 chunking to Python + TS embed_batch
- Moved all inline imports to top level in main.py
- Cleaned up GraphStoreError dead code from exceptions.py

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

mem0-ts/src/oss/tests/memory.crud.test.ts

@@ -22,18 +22,23 @@ jest.mock("../src/llms/openai", () => ({
       .fn()
       .mockImplementation(
         (messages: Array<{ role: string; content: string }>) => {
-          const hasSystemRole = messages.some((m) => m.role === "system");
-          if (hasSystemRole) {
-            return JSON.stringify({ facts: ["stored fact"] });
-          }
+          // V3 pipeline: single LLM call with additive extraction prompt.
+          // Extract the user input from the prompt to produce unique memories.
+          const userMsg = messages.find((m) => m.role === "user");
+          const content = userMsg?.content ?? "";
+          // Pull the text between "## New Messages" and the next "##"
+          const newMsgMatch = content.match(
+            /## New Messages\n([\s\S]*?)(?=\n##|$)/,
+          );
+          const extracted = newMsgMatch
+            ? newMsgMatch[1].trim()
+            : "stored fact";
           return JSON.stringify({
             memory: [
               {
-                id: "new",
-                event: "ADD",
-                text: "stored fact",
-                old_memory: "",
-                new_memory: "stored fact",
+                id: "0",
+                text: extracted,
+                attributed_to: "user",
               },
             ],
           });
@@ -42,9 +47,15 @@ jest.mock("../src/llms/openai", () => ({
   })),
 }));
 
+const mockEmbedding = new Array(1536).fill(0.1);
 jest.mock("../src/embeddings/openai", () => ({
   OpenAIEmbedder: jest.fn().mockImplementation(() => ({
-    embed: jest.fn().mockResolvedValue(new Array(1536).fill(0.1)),
+    embed: jest.fn().mockResolvedValue(mockEmbedding),
+    embedBatch: jest
+      .fn()
+      .mockImplementation((texts: string[]) =>
+        Promise.resolve(texts.map(() => mockEmbedding)),
+      ),
     embeddingDims: 1536,
   })),
 }));