feat: persistent memory layer

[pkarw]

Add a vector store to save the execution context between the calls. With the persistent storage, we will be able to

• implement async calls (ag continue the flow after Slack/Email message from where it paused)
• implement a Knowledge repository for the agent to use and aggregate
• optimize the context window (RAG)

It should be dependency-less.
It should be abstract - so one can switch the driver within the Team configuration.
It probably should provide a toolkit like @tools/vectorstore -> add, find etc.

TBD: to select a vector store provider for the initial use, options:

• https://github.com/tursodatabase/libsql SQLite
• https://github.com/asg017/sqlite-vec
• https://js.langchain.com/v0.1/docs/modules/data_connection/vectorstores/ - maybe as an option, as it's too complex for a default (?)

[pkarw]

Should be implemented both ways:

• vector store (described above)
• simple KV tool - memoryGet, memorySet + configurable backend

Wdyt @grabbou?