Optimize storm cell detection performance

[danielway]

Scope

Optimize the accepted sweep-observation, volume-cell, and temporal-tracking pipeline without changing documented results. Execute the pipeline in a Web Worker so detection never blocks UI frames.

Baseline and target

• Benchmark the shared curated corpus from Improve storm cell detection quality #164 on a documented representative mid-range laptop/browser.
• Process each newly completed representative volume in under 250 ms.
• Record CPU time, peak working memory, and allocations for each pipeline stage.

Requirements

• Cache by input volume/sweep identity and detection parameters, including valid empty results.
• Do not repeat unchanged work on UI frames.
• Incremental live updates process only newly available inputs where feasible.
• Worker results are versioned so stale results cannot overwrite newer state.

Acceptance criteria

• The documented benchmark meets the under-250-ms completed-volume target.
• Main-thread frame work does not perform detection/aggregation/tracking computation.
• Outputs match accepted fixtures within documented floating-point tolerances; intentional differences are reviewed and recorded.
• Empty-result caching, stale worker results, incremental updates, and parameter invalidation have regression coverage.
• Before/after benchmark results and measurement method are documented.

Depends on #164, #178, and #179. Parent: #33.