Add zoomable multi-window tab canvas - #3

Summary

• replace the single-window grid with a continuous zoomable map of every normal Chrome window
• open centered on the invoking window, with pan/zoom and a deliberate Fit action for the all-window overview
• search across all windows without changing spatial layout; nonmatches dim and keyboard-selectable results stay available
• keep Position, Domain, and Recent sorting scoped independently to each real window
• add tab drag/drop within and between windows, including drop-position handling
• add session window labels, cross-window activation, and live window event refreshes
• preserve Reading List, close, recency/domain colors, status metadata, dark mode, origin return, and close-target stability
• crossfade detailed controls at readable zoom while reserving their geometry so cards never jump
• document the deferred AI experiment and product lessons in docs/AI_FEATURE_LEARNINGS.md

Product decisions

• AI controls are not included in the extension UI
• the invoking window is the initial camera target, but the camera is not permanently locked
• search highlights within the stable map instead of filtering windows or resizing cards
• rich tab details fade with zoom using separate enter/exit thresholds to prevent threshold flicker
• user-created window labels use extension session storage because Chrome does not expose its native window name through the extension windows API

Validation

• npm run typecheck
• npm test — 42 tests passing
• npm run build
• verified the packaged build contains manifest.json, service-worker.js, and the Eagle page/assets