Clarify Agent Memory Protocol and Tooling

During the first live test of the memory core, the agent failed to follow the "save-then-respond" protocol due to two issues:

1. A misinterpretation of the procedural order in AGENTS.md, where the memory check was not performed during initialization.
2. A lack of clarity in the addMemory.mjs script's command-line interface, leading to several failed attempts to save the session history.

This ticket covers the documentation and process improvements needed to prevent these failures in the future.

Acceptance Criteria

1. The .github/AGENTS.md file is updated to move the memory server check into the "Session Initialization" section.
2. The buildScripts/ai/addMemory.mjs script is updated with clear documentation for its command-line options, especially --thought and --session-id.