Visualizer compaction and optimization

[Amos-Rai-KEYS]

PR Summary: Visualizer Compaction and Performance Optimization

Overview

Adds large-fabric support to the infrastructure visualizer: server-side layout precomputation, a compressed-view ladder for large topologies, on-demand "rack/pod" drill-down views, and rendering optimizations for large graphs. Also includes a couple of small fixes and README cleanup.

Key changes

1. Precomputed layout (visualize.py)

• New Visualizer._compute_layout assigns hierarchical level + x/y position to every instance node server-side (BFS from hosts + barycenter ordering), so the browser no longer needs to run vis.js's hierarchical layout/physics engine to open large graphs.
• Edges now carry a count (parallel-link count) alongside the existing label/width.

2. Compressed view ladder (visualize.py, navigation.js)

• For fabrics with more than 128 hosts (COMPRESS_MIN_HOSTS), _generate_compressed_views builds a ladder of infrastructure_compressed_N.json views, each merging nodes with identical connectivity one step further than the last, ending in one node per tier.
• Groups are semantically labeled from data: a bottom-tier group whose members all uplink to one switch → rack; several racks merged → pod; everything else → group.
• Frontend adds a Compressed toggle button + slider in the header (navigation.js) to switch the infrastructure view to a chosen rung, only enabled/visible when on the top-level infrastructure view.

3. Rack/pod drill-down (navigation.js, data.js, app.js)

• Clicking a rack/pod group node (instead of a switch group) opens a synthesized "rack view": the group's real member servers plus switches they uplink to, built on demand from infrastructure.json and laid out client-side (layoutSubgraph, a JS port of the Python layout).
• A hop slider controls how much surrounding fabric is included; capped at 250 nodes (RACK_MAX_NODES) to stay responsive, with a toast shown when capped or refused.
• Groups too large to open (highest compression rungs) show a "not-allowed" cursor and toast instead of silently doing nothing.

4. Large-graph rendering mode (network.js, data.js, app.js)

• Graphs ≥300 nodes or ≥1000 edges automatically switch to a simplified rendering mode: no text stroke/shadow, straight (non-bezier) edges, hover effects disabled — these are the slowest canvas primitives on pan/zoom.
• Manual override checkbox in the controls panel; optionsForData() centralizes picking precomputed vs. hierarchical vs. large-graph vis.js options.

5. Misc frontend fixes

• controller.js: slider handlers (font/node size, spacing) are now debounced (120ms) instead of firing on every pixel of drag; precomputed-view spacing sliders rescale positions directly instead of re-running vis layout.
• app.js / render(): uses a Map lookup (nodeById) instead of Array.find for click/hover node lookup; skips the stabilizationIterationsDone physics dance entirely when a view has no physics.
• search.js: node search result colors now use CSS theme variables instead of hardcoded hex, so results respect dark/light theme.
• visualize.py: removes two now-unnecessary lines from the pipeline (pipeline fix commit).

6. Docs

• README.md / docs/src/cli.md: document the compressed view, rack/pod behavior, and --hosts guidance.
• Removes src/infragraph/visualizer/README.md (407 lines) and a couple of lines from the root README in the final commit — worth confirming this was an intentional retirement of that file rather than an accidental deletion, since there's no equivalent content added elsewhere in the diff.

Files touched

visualize.py, navigation.js (+341 new), network.js (+42 new), data.js, app.js, controller.js, filters.js, search.js, style.css, index.html, README.md, docs/src/cli.md.