Add memory_batch MCP tool for bulk memory_entries writes

[martineserios]

Problem

agentdb_batch writes to the episodes table, not memory_entries. There is no batch tool for memory_entries, forcing users to either:

1. Call memory_store sequentially (slow — serialized by the MCP server, ~2-3h for 1200+ entries)
2. Write directly to SQLite using ruflo's own deps (our current workaround)

Request

Add a memory_batch MCP tool that bulk-inserts into memory_entries with proper embeddings, similar to how agentdb_batch works for episodes.

Suggested interface

{
  "tool": "memory_batch",
  "params": {
    "entries": [
      {
        "key": "knowledge:dimension:doc:section",
        "value": "content text",
        "namespace": "knowledge",
        "tags": ["source:brana-knowledge", "type:dimension"],
        "upsert": true
      }
    ]
  }
}

Context

• memory_store targets memory_entries table — correct table, but no batch support
• agentdb_batch targets episodes table — has batch support, but wrong table
• guidance recommends embeddings_batch, memory_import, memory_export, memory_compact — but none are wired as MCP tools
• Our workaround: Node.js script using ruflo's own better-sqlite3 + @xenova/transformers to write directly to memory_entries with 384-dim embeddings (1247 sections in 88s)

Related

• Table split between memory_entries and episodes is undocumented — users expect agentdb_batch to work for memory storage