Add File-Based Persistent Memory

PROJECT CONTEXT:
My agent forgets everything between sessions and I keep being told I need a vector database. Practitioner reports and memory-system analyses show plain files (markdown/JSONL) cover most agent memory needs with full inspectability.

GOAL:
Implement file-based memory: a project memory file the agent reads at start and updates through a guarded tool, all inspectable in a text editor.

REQUIREMENTS:
- A MEMORY.md (or agent-memory directory) per project with defined sections
- Memory loaded into context at session start
- A memory_update tool the agent uses to add or revise entries
- Human-readable format: a user can read and edit the memory directly
- Size discipline: per-section caps with a consolidation routine

CONSTRAINTS:
- No databases, no embeddings in v1 - files only, versioned with the project
- Every memory write goes through the tool so writes are validated and logged

DEFINITION OF DONE:
- Facts recorded in one session are used correctly in the next
- The memory file remains readable and sensibly organized after 20+ writes
- User edits to the file are picked up on next session start
- Sections stay under their caps via consolidation, not silent truncation

COMMON FAILURES TO AVOID:
- Building embedding infrastructure for what a 200-line markdown file handles
- The agent appending duplicates because it never reads before writing
- A memory file that decays into an unstructured dump nobody trusts
- Writes scattered through code instead of one validated tool path