feat: add /nemp:foresight command for intent-based memory filtering

Scores all memories against a prompt's intent and returns only
relevant ones, replacing full memory dumps with a targeted subset.

- 13 domain detectors (auth, database, api, frontend, backend,
  testing, deployment, payment, design, state, security,
  architecture, feature) with signal word tables
- 4-component scoring: domain match (0.40), keyword match (0.30),
  key name match (0.20), recency (0.05)
- Core key floor of 0.30 ensures stack/framework always included
- Auto top_k from intent level: high=12, medium=8, low=5, continuation=20
- --top and --threshold flags for manual override
- Token savings % shown in output stats
- Logs each call to .nemp/access.log
- Also updates CLAUDE.md to reflect skills/ dir added in v0.2.0

Author: Sukin Shetty

CLAUDE.md

@@ -1,6 +1,6 @@
 ## Project Context (via Nemp Memory)
 
-> Auto-generated by Nemp Memory. Last updated: 2026-02-11 14:15
+> Auto-generated by Nemp Memory. Last updated: 2026-02-17 00:00
 
 ### Project Info
 
@@ -16,7 +16,7 @@
 |-----|-------|
 | **stack** | Claude Code Plugin — Markdown commands, JSON storage, agent tracking, access logs, token compression |
 | **storage-format** | JSON files — .nemp/memories.json (project), ~/.nemp/memories.json (global). Plus access.log audit trail and MEMORY.md index. |
-| **project-structure** | commands/ (16 .md files), .claude-plugin/ (plugin.json, marketplace.json, hooks/), .nemp/ (config, memories, access.log), assets/ (logo/, images/, demos/) |
+| **project-structure** | commands/ (16 .md files), skills/ (nemp-memory/SKILL.md), .claude-plugin/ (plugin.json, marketplace.json, hooks/), .nemp/ (config, memories, access.log), assets/ (logo/, images/, demos/) |
 
 ### Features