perf: decide between removing the worker thread and splitting the postMessage envelope

[msallin]

The per-tick `worker.postMessage(srcPaths)` structured-clones the entire path snapshot, including `activeRoute.waypoints`. Two competing ways to eliminate that cost:

• Option A — remove the worker. Call `calcs()` inline on the main thread. Eliminates IPC overhead and the worker's V8 heap. Simpler if inline calc latency on a worst-case tick stays well under one delta interval.
• Option B — split the envelope. Keep the worker but split `srcPaths` into a stable envelope (`waypoints` + `waypointsVersion`, sent only on bump) and a tick envelope (vessel position, COG, SOG, etc., sent every tick). Worker holds the last stable envelope by reference.

A/B both options against the dispatch benchmark from #24 on Pi 3-5 hardware and pick one.

[panaaj]

I vote for Option A if the IPC overhead is the actual issue.

[msallin]

Draft PR for Option A: #32. Removes the worker thread, runs calcs() inline.

The PR description has bench numbers from a primed dispatcher bench (master with mocked postMessage vs branch). Caveat noted there: master's bench numbers are artificially cheap because the worker boundary is mocked away — a real-worker bench would tighten the comparison but is out of scope.

Open as a draft so the trade-off (main-thread CPU per tick goes up vs structured-clone + IPC + separate worker heap go away) can be discussed before commit. No PR opened for Option B yet.