fix(dashboards): Stop chart drag-to-zoom selecting the wrong time range

[gggritso]

Dragging to zoom on a dashboard time-series widget could snap the global time filter to the wrong (collapsed) range — e.g. a ~2h selection landing on ~1 minute.

Two related bugs in Dashboards time series chart navigation

1. Dragging to zoom creates weird opaque rectangles
2. Drag-to-zoom seems to zoom to incorrect range if many zooms are done without reloading

It looks like the root cause is apache/echarts#21661, in which re-mounting a chart reinitiates the zoom toolbox, which weirdly re-fires old zoom events? Truly bizarre. To guard against this, allowing notMerge=false makes sure that ECharts doesn't do a full re-init of the toolbox, it just updates it if it changed (it didn't), which prevents this weird re-initialization loop.

I worked through this with Claude for a while, and I'm convinced that the solution works. Since notMerge is able to cause broken UI, I removed that prop completely.

This is the same root cause as LOGS-901.

Fixes DAIN-1753

Investigation notes

Verified live with temporary console instrumentation on the repro dashboard (single time-series widget)

Repro: drag-zoom (~2h) → dashboard Cancel → drag-zoom again → second selection collapses to a wrong, tiny range and the selection rectangle looks more opaque.

Two theories were disproven by the logs, not assumed:

• Orphaned ECharts "cover" SVGs stacking (the sibling LOGS-901 hypothesis): the brush controller's tracked covers always equalled its mounted covers (0/0 or 1/1) — no orphans.
• Timezone offset: would corrupt the first drag and shift by whole hours, not collapse the second drag to ~1 min.

What the logs actually showed: a single drag fired handleDataZoom 3–6 times, each with a different from: toolbox-feature_NNN uid climbing by +26 and a drifting range (10 → 126 → 126 → 126 → 12 min); the last write won and it was the stale collapsed one.

• Not leaked listeners: echarts-for-react unbindEvents()/bindEvents() on every update, and the events had different from/ranges (a single event to N stale listeners would be identical N times). These are N genuinely distinct dispatches.
• The +26 uid jump = the toolbox feature being rebuilt from scratch each time = full notMerge re-init.
• Shape is a feedback loop: drag → setPeriod → URL/page-filter change → widget re-render → setOption(notMerge:true) full re-init → toolbox dataZoom re-emits a dataZoom → setPeriod → … until it settles on a drifted/collapsed range.

Fix confirmation: with notMerge={false}, each drag fires handleDataZoom exactly once, the from uid stays stable, and both drags land the correct ~2h range (117 / 119 min).

Upstream: apache/echarts#21661 (open, "pending") is the same regression; #21644/#21669 are related. No merged upstream fix exists, so this is a client-side mitigation.

Blast radius: dashboards, insights, and all explore time-series charts now merge (they shared the latent bug; metrics/spans were already on notMerge={false}).

[linear-code]

DAIN-1753

[github-actions]

Story previews

Preview the stories changed in this PR on the Vercel deployment:

• app/views/dashboards/widgets/timeSeriesWidget/timeSeriesWidgetVisualization.stories.tsx

_{Preview deployment: https://sentry-le5uzmcc4.sentry.dev}

[gggritso]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 8047a99. Configure here.}