docs: refresh all docs to v0.17.2 + v0.18-alpha state

- COMPARISON.md: replace v0.12 benchmark table with 14-bench FTS vs
  Full pipeline (avg 0.615→0.647, +5.2%) + multiturn agent 172q
  (Qwen3.5-27B, 81.4%) + MuSiQue known limit (−0.294 vs HippoRAG2)
- CONCEPTS.md: correct edge weights (MENTIONS 0.5→0.8, add
  EXTRACTED_FROM/DEPENDS_ON) and NodeKind table (no DOCUMENT kind,
  ENTITY is reused for documents); clarify 3-tier ProfileGenerator
  with explicit optional LLM tier; update §12 limits + roadmap
- GUIDE.md: rewrite §7 benchmarks for v0.17.1 (single-shot hits +
  multiturn agent table vs v0.13 baseline); FAQ → v0.17.2 stable
- ARCHITECTURE.md: full rewrite to v0.17.1 architecture (170→500+
  lines, new facade diagram + 14-bench numbers)
- README.md: fix broken GUIDE.md §7 anchor, add PLAN-v0.17 +
  PLAN-v0.18 links, update PyPI stable to v0.17.2
- ROADMAP.md: restructure — v0.16/v0.17 moved to historical, add
  active v0.18 track (α1-α3), carry-over section (C1-C5), measured
  negatives table (6 approaches), v0.17.2 entry (Apache-2.0 +
  agent_loop fix), design principle #9 added

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: SonAIengine