Scott/link click chart select

[sagrimson]

Overview

When a user clicks a link on a card to create a chart, the atlas now intelligently handles the case where the chart already exists — instead of duplicating it, the camera pans to bring the existing chart into view alongside the source card. Additionally, the pan/zoom behavior when focusing a card pair has been refined: if the target chart is already fully visible no camera animation occurs, and when panning is needed the camera zooms out to frame both the source card and the chart together rather than simply centering on the chart.

Changes Made

• Added focusTarget state to the atlas store to track which card pair (source + target) should be focused
• When cardRegisterChart is called for an already-existing chart (or a generic chart already showing the same variable), set focusTarget to pan to it instead of creating a duplicate
• Created register_card_focus.ts — a new atlas side-effect that watches focusTarget and moves the camera to frame both the source and target cards (using zoomToBounds with a capped zoom level), or waits for the shape to appear if it hasn't been created yet
• Added FOCUS_ZOOM_DISPLAY_CAP config constant (125%) to prevent the camera from zooming in too far when both cards are close together
• Exported resolveGrid and canFitWithinZoomCap from register_card_placement.ts for reuse in the focus logic
• Refactored duplicated node-removal logic in queryCancel and nodeDismissFollowUp into a shared removeNode helper

[gemini-code-assist]

Code Review

This pull request introduces a card focusing mechanism that adjusts the camera zoom and position to keep both the source and target cards in view when a focus target is set. It integrates this behavior into the AtlasProvider and updates the AtlasStore to manage the focus target state, while also refactoring the store to reduce duplication using a new removeNode helper. Feedback on these changes suggests correcting a comment mismatch regarding the zoom cap percentage, simplifying shape ID handling by casting directly to TLShapeId instead of redundantly parsing the prefix, and refactoring the asynchronous afterCreate handler registration to prevent potential reference errors or memory leaks.

[beets]

Thanks for the update Scott!

The code looks great, but I have some concerns around the zoom behavior. My expectations when interacting is the zoom / scroll behavior will be the minimal movement required to get the source and selected card in frame, not the midpoint of the two.

In this example, I expected the canvas the selected card to be on the bottom left, not middle of the screen (which is quite different from the behavior when the right card is selected).

I wonder if the behavior should be updated to also account the currently selected card or the last added card, and we compute the midpoint of all three. Or other ideas to minimize the scroll.

Screen.Recording.2026-07-21.at.10.25.09.PM.mov

Could you upload this PR to dev so we can play with it more? I've also just approved all open PRs, so it would be nice to get a snapshot of all the latest code there. /cc @miss-o-soup

[sagrimson]

@beets thanks for that feedback it made me rethink & it feels much nicer now! I updated so the minimum possible animation is used to get the newly selected card on screen -- It's live on .foo so let me know what you think!