fix(web): project settings can set a checkout's grouping rule again#23641

Problem

With related checkouts grouped in the default sidebar, Settings → Project → Checkouts had no per-checkout grouping control. A checkout could not be split into its own project there, or restored to the global rule after being split.

Change

Restore the existing four grouping choices on every checkout, including single-checkout projects. Persist changes through the client-settings update queue using the same override keys as the legacy dialog; inheriting removes only the selected override. Existing navigation follows the checkout when its current group empties.

Save failures display an error toast and retain the saved selection. Accessible names include the environment label and full ID. Checkout controls stack at narrow widths; a lone checkout keeps its removal action in Danger. The legacy dialog shares the labels and override helper.

Scope and approval

Fixes #16445

Triaged bug via the automated issue flow: #16445. This restores an existing client setting removed from project settings by #10639; it adds no contract, server, schema or native-client behavior.

Verification

At f822473:

• Focused ProjectSettingsPanel logic tests: 3/3 passed. Targeted lint, formatting and web typecheck passed. All32 reported CI checks passed; both independent full code reviews returned MERGE. Macroscope approved and CodeRabbit reports no actionable findings.
• Real-app checks cover grouping/inheritance, preservation of other overrides, scoped split/return navigation, two environments with the same path, keyboard interaction, reduced project permissions, legacy Save/restore and both persistence-rejection messages.
• Independent corrected proof audit: SUFFICIENT. All112 comparison counts/masks were independently verified. One exact-path head rebuild matched with0 changed pixels at both thresholds.

Evidence

Complete matrix with every embedded comparison and captioned before/after recording · Reproducible fixture and stills · Source and captioned recordings.

The full matrix exceeds the PR body limit; representative rows are embedded below. Actual feature base is f054d1b; failure-feedback comparisons use the separate pre-toast 33b1806. Every after is f822473.

T3 preview, Chrome154/Linux, no device emulation, DPR2,16px root font;1440×900 and390×844 CSS viewports, light/dark. Matched pairs use identical synthetic data, origin, frozen Date, numeric scroll and focus. Wide PNG transport downsamples to1280×800; narrow PNGs are780×1688. Amplified masks show every pixel at>0 and>8/255 (3.137%, not8%). Explicitly labelled bottom-anchor views use different scroll offsets and are backed by common-coordinate pairs.

Captioned scoped round trip: select Keep separate, observe automatic navigation while retaining scope, then Use global default and observe automatic return. Captions distinguish source checkpoint/wall-clock time from compressed playback. The base counterpart shows the absent control.

Visible changes: the helper and grouping selectors add height; Remove controls shift, nested rows stack, and narrow paths/default labels truncate. The failure toast overlays the upper settings content at narrow width. Wide legacy-dialog and breadcrumb glyph raster differences are disclosed per row; no claim of zero unrelated pixel changes is made. Historical invalid/contaminated comparisons are excluded, with provenance retained in the linked reports. Web content used by desktop is covered; no native Electron-shell change or mobile grouping override was introduced. All owned capture/audit servers are stopped.

Model and harness: Claude through Claude Code; GPT-6.1-Sol through Codex in T3 Code.