perf(INP/desktop): deckgl layer-toggles header p75 11.7s — worst desktop target by two orders (#4537 Track D)

## Problem
`deckgl-layer-toggles > .toggle-header` is the single worst desktop INP target by p75: **11,744ms (n=7)** in the 2026-07-09 Sentry device split — two orders of magnitude past the next offender. Small n, but 7 near-12s interactions is a real defect, not noise. Surfaced during the #4537 refresh; currently unowned.

## Where
The layer-toggles panel collapse/expand header — `src/components/DeckGLMap.ts` (and `GlobeMap.ts` mirror). Hypothesis to test FIRST: the expand/collapse triggers a synchronous deck.gl layer rebuild or a full `updateLayers()` commit in the click frame (cf. closed #4558 which chunked `updateLayers` with yieldToMain — this path may bypass that).

## Diagnose first
`node scripts/measure-dashboard-render-axis.mjs "https://www.worldmonitor.app/dashboard" --interact "selector:.deckgl-layer-toggles .toggle-header" --cpu-throttle 4 --json` at DESKTOP viewport (`--width 1365 --height 768`) — worst-event-scoped phases + symbolicated top events name the culprit (prod keeps function names).

## Acceptance
Sentry desktop `webvital:inp` bad-event rate for this selector → ~0; DebugBear desktop `inpSelector` row p75 < 500ms. Part of #4537 Track D.

Start with src/components/DeckGLMap.ts and its GlobeMap.ts mirror, then run scripts/measure-dashboard-render-axis.mjs against the dashboard at the stated desktop viewport and interaction. Check whether expanding the layer-toggles header triggers a synchronous deck.gl rebuild or full updateLayers commit, using the symbolicated event phases to identify the culprit. Done means the Sentry bad-event rate is approximately zero and DebugBear desktop inpSelector p75 is below 500ms.

Written by the indexing model from the issue text.