fix(agent-framework): give each add_memory call its own document

[harshitgavita-07]

Problem

SupermemoryTools.add_memory in agent-framework-python passes custom_id=self._connection.custom_id (conversation_<id>) on every call.

In Supermemory a stable customId identifies one document, and re-adding under it updates that document (see the customId notes in the ingest docs). So with this tool:

• the second add_memory call in a conversation replaces the first remembered fact instead of adding a new one, and
• every tool-saved fact shares an id with the conversation transcript that the middleware and context provider save under the same connection.custom_id, so they overwrite each other.

supermemory_openai's add_memory does not pass a custom_id, so each fact is its own document there. This makes the agent-framework tool match.

Fix

Drop custom_id from the add_memory tool call. connection.custom_id is unchanged and still used for conversation capture.

Test

New TestAddMemoryDocumentIdentity records the arguments of two add_memory calls and asserts neither reuses connection.custom_id. It fails on main and passes with this change. Full package suite: 59 passed (58 existing plus the new one).

I did not run this against the live API. The overwrite behavior comes from the documented customId semantics, not from a live reproduction.