refactor: redesign notebook cell dimensions and responsive views

[emrberk]

Summary

• Redesign notebook cell sizing and responsive views so users can independently size the editor and results, preserve their preferred layout across viewport changes, and use consistent resize and visibility controls.
• Replace grid-height and maximized-view agent semantics with independent editor_height, result_height, and view controls.
• Report the effective responsive view and compact/wide tier while preserving the preferred view intent.
• Add a per-cell Stop control for cancelling initial query runs and chart fetches.
• Make chart zoom controls responsive to the rendered data density and available width, while keeping Reset Zoom available across viewport changes.
• Treat hiding a result as an explicit discard action that clears the current result and saved snapshot while preserving the preferred layout for future runs.
• Preserve run history and copy matching saved result snapshots when duplicating notebooks, without switching the active tab.
• Return faithful JSON previews and preserve semantic dimensions in snapshots.

Validation

• yarn typecheck
• yarn test:unit (2,160 tests across 95 files)
• focused notebook snapshot/tool/controller tests (252 tests)
• yarn build
• targeted ESLint and git diff --check

Deferred follow-up

• Update mcp-server-questdb for the new notebook tool contract.
• Bump the MCP server version.
• Update EXPECTED_MCP_VERSION in this repository after that release.

[emrberk]

PR #606 Review

Reviewing PR #606 at level 3

Reviewed head: a36aa4e812b15270f237aaddb27ab6f3d2031d34
Base revision: 72ca5487b263fba39ecd2e9221066b03535b4c75

The title and description follow the repository conventions.

Issues

Issue ID	Issue name	Category	Severity	Location	Net impact	Evidence	Description	Steps to reproduce	Suggested fix
#1	Refresh rerenders entire cell	Performance & rendering at scale	Moderate	in-diff — src/scenes/Editor/Notebook/cells/Cell.tsx:164	Multi-statement dashboards spend up to 141 ms rerendering one cell.	Playwright React profiler: the same eight-statement refresh produced 13 renders/44.7 ms at head a36aa4e, versus 2 renders/6.5 ms at base 72ca5487.	CellInner subscribes to every raw refresh-engine state object, although grid cells do not consume the subscribed value. Eight statements add eleven full-cell renders; 32 statements produced 37 renders and 141.4 ms of cell work. This repeats at the two-second auto-refresh minimum and compounds across dashboard cells.	1. Create an eight-statement grid cell. 2. Run it, then refresh the unchanged results. 3. Count CellInner commits: head consistently produces 13 versus 2 at base.	Subscribe to a selector-filtered fetching boolean, suppress unchanged values, and avoid installing the subscription for grid cells—for example, isolate the draw Stop gate in a small child.
#2	Divider collapses tall cell	Accessibility & UX	Moderate	in-diff — src/scenes/Editor/Notebook/cells/useCellResizeOrchestration.ts:111	Tall split-cell users lose 2,300 px in one divider interaction.	Scratch Vitest at head a36aa4e: 4,800 px becomes 2,500 px. Base End path is N/A because keyboard resizing is a new surface; the equivalent base mouse partition preserves the total.	With both panes at the permitted 2,400 px maximum, the middle divider advertises a 4,700 px maximum. Pressing End partitions the cell as 4,700/100, after which the independent pane clamp converts it to 2,400/100, unexpectedly collapsing the total height.	1. Set a split SQL cell's editor and result heights to 2,400 px. 2. Focus the middle divider. 3. Press End. 4. Observe the allocation fall from 4,800 to 2,500 px.	Derive divider bounds so both panes stay within their individual maxima: clamp the requested top between sum - MAX_PANE_HEIGHT_PX and min(MAX_PANE_HEIGHT_PX, sum - minBottom). Add the 2,400/2,400 boundary test.
#3	Stop control untested	Test review & coverage	Moderate	in-diff — src/scenes/Editor/Notebook/cells/CellStopButton.tsx:28	Notebook runners lack regression coverage for the new cancel control.	Exact test search at head a36aa4e found no component or data-hook matches; lower-layer cancellation tests passed 141/141. Base is N/A because this is a new surface.	Existing tests verify cancellation helpers and the refresh engine directly, but none mounts the new control or verifies its visibility, grid/chart routing, click dispatch, removal, or focused-unmount handoff. Those integration points could regress while every current test remains green.	1. Hold a notebook cell's first /exec request. 2. Assert that the Stop control appears and click it. 3. Assert cancellation and button removal. 4. Repeat for the initial chart-fetch branch.	Extend the existing delayed-/exec Cypress notebook flow to cover grid and chart cancellation, including focus handoff.

Validation

• yarn typecheck: passed
• yarn build: passed
• yarn lint: passed
• yarn test:unit: passed — 95 files, 2,160 tests
• git diff --check: passed
• Browser rung: the responsive chart zoom/reset path passed with Playwright against the no-auth localhost:10000 server. The repository Cypress helper hard-codes an authenticated login, so that scenario was executed directly in the browser.
• Temporary review files and worktrees were removed. Pre-existing workspace changes were left untouched.

Summary

• Verdict: approve with comments — address wip test for error range #1–disabled buttons #3, but none is blocking.
• Correctness gate: passed — no admitted Critical findings.
• Test gate: passed — 1 admitted Moderate coverage gap and 0 Critical coverage gaps.
• Findings: 3 in-diff, 0 out-of-diff. The level-3 cross-context inventory was rechecked and the remaining findings are contained within the changed surface.
• Severity: 0 Critical, 3 Moderate, 0 Minor.
• Adjacent findings: none.

[github-actions]

Web Console deploy preview

Preview	Commit	Logs
https://pr-606--web-console.netlify.app	48bfc80	build log