feat: add human-friendly architecture dashboard

[kluacisco888]

Summary

• Adds dashboard.py that generates a single-page HTML architecture overview from graph.json
• Merges fine-grained Leiden communities into 5-10 macro-modules using hierarchical clustering + directory-aware grouping
• Renders an interactive D3 force graph (clickable nodes), module cards, and PageRank-based reading order

Motivation

The current outputs (GRAPH_REPORT.md, graph.html) are optimized for AI consumption. For human team members — especially during onboarding — a 453-line report and a 1953-node force graph are overwhelming. This dashboard provides a simplified, human-first view of the same data.

What's included

New: graphify/dashboard.py (~500 lines)

• merge_communities() — groups fine-grained communities into 5-10 macro-modules by cross-community edge density + directory structure
• _macro_metadata() — extracts node count, file count, key files, top nodes, inter-module dependencies
• _infer_module_name() — auto-names modules from directory structure and node labels
• build_description_prompt() — generates a prompt for the host LLM to produce one-line business descriptions
• _reading_order() — PageRank-based recommended reading order for new team members
• generate_dashboard() — public API, accepts optional names and descriptions overrides
• Single-file HTML output with D3.js force graph, dark theme matching graph.html

Modified: graphify/__main__.py

• New graphify dashboard CLI command with --graph and --output flags

Modified: graphify/skill.md

• New Step 6c: auto-generates dashboard during /graphify pipeline

Usage

graphify dashboard
graphify dashboard --graph path/to/graph.json --output path/to/dashboard.html

from graphify.dashboard import generate_dashboard
result = generate_dashboard("graphify-out/graph.json",
    names={0: "Auth Module", 1: "Data Pipeline"},
    descriptions={0: "Handles user authentication", 1: "ETL pipeline"},
    project_name="My Project")

Design decisions

• No new dependencies — uses only NetworkX (already required) and D3.js (CDN)
• Generic — works on any project's graph.json, no domain-specific logic
• LLM-optional — auto-infers names from paths/labels, LLM descriptions are an enhancement
• Dark theme — matches existing graph.html aesthetic