[FEATURE] Per-folder notebook memory (notes_get/notes_commit) for the agent

Summary

Add per-folder notebook memory so the agent persists what it learns across tasks and reuses it instead of re-exploring code.

What

• Each folder can carry a .note.yaml with a folder summary, per-file entries, and cross-file relations, grounded in based_on source fingerprints with freshness badges (✓/⚠/✗).
• notes_get reads the ancestor-chain memory for a path or keyword-searches all notebooks.
• notes_commit writes summaries/entries/relations with diff approval (write-notes permission), never auto-approved.
• A concreteness gate rejects summaries that name no concrete symbols, so saved notes are usable mid-task.
• NotebookAttach prepends the relevant per-file note to read/edit/write/apply_patch output on first touch (opt-out via experimental.notebook.attach = false).
• The injected digest tells the model it only samples memory and to call notes_get before answering general project questions.
• Save-learnings nudge after exploring without saving.

Motivation

Agents re-read code they have already mapped; a lightweight, source-grounded per-folder memory with freshness badges avoids that and keeps the map trustworthy.

Locate the notes_get and notes_commit entry points and the NotebookAttach implementation first. Trace how folder memory, source freshness, write-notes approval, concreteness checks, and tool-output attachment are represented. Done means per-folder notes can be read and safely committed, relevant notes attach as specified, and the opt-out and save-learnings behavior work.