Update memory docs for project working memory

[biswaroop1547]

Summary

• Rework the Memory page around the shipped model (premAI-io/premapp-backend#668): global preferences, project context, and automatic project working memory that lets a new chat pick up where the project left off.
• Replace stale mechanics: memory loads with every message rather than once per chat, and background learning runs after each reply instead of at a token threshold, saving durable items as auto-saved, deduplicated entries and dropping anything secret-looking.
• Add a "Reading and correcting it" section: preferences and project context are plain markdown in the workspace, and working memory can be reset on request.
• Retune the Going deeper audit prompt to the current layers.
• Add Cowork to the spell-check dictionary (same one-line fix as Document steering and the follow-up queue #28; main is red on cspell): merges cleanly whichever lands first.

Validation

• mint broken-links
• cspell clean repo-wide

🤖 Generated with Claude Code