refactor: simplify continual learning — single script, two-tier memory, auto-init

[thegovind]

Follow-up to #159. Redesigns the continual learning hook for simplicity.

What Changed

Before (v1): 3 separate scripts, 4-step install, flat memory, manual setup
After (v2): Single script, 1-step install, two-tier memory, auto-init

Install

cp -r hooks/continual-learning .github/hooks/

One step. Everything auto-initializes on first session.

Key Design Changes

Single Script

learn.sh handles all lifecycle events via argument — no multi-script complexity:

learn.sh sessionStart   → load learnings
learn.sh postToolUse    → track tool outcome (3ms)
learn.sh sessionEnd     → reflect, persist, compact

Two-Tier Memory

Scope	Location	Purpose
Global	~/.copilot/learnings.db	Tool patterns — follows you everywhere
Local	.copilot-memory/learnings.db	Repo conventions — stays with the project

Compaction

• Old low-value learnings decay (60-day TTL, low hit count)
• High-value learnings persist indefinitely (ranked by hit_count)
• Tool logs pruned after 7 days

Auto-Init

DBs and directories are created on first run. No manual mkdir, chmod, or setup steps.

Files Changed

• hooks/continual-learning/learn.sh — new single script (replaces 3 scripts)
• hooks/continual-learning/hooks.json — updated to call learn.sh with event args
• hooks/continual-learning/README.md — concise, beads-style docs
• .github/skills/continual-learning/SKILL.md — updated to match new design