fix(ui): every tooltip answers the pointer, through one class handler

[Shman4ik]

What and why

#332 found that a TextBlock or panel with no Background is not hit-testable, glyphs included: 0 of 4,536 points over the status text reached the text block, so its tooltip never opened. It patched the status bar only.

A new headless walk (TooltipReachTests) hit-tests every tooltip-bearing element in every Scenarios.All window. On main it finds 203 failures, about 50 distinct elements:

• history rows (statement and run detail) and saved-query rows
• connection list endpoints
• Activity and Slow Queries query text
• Database Overview's cache-hit tiles, the Security window's "connected as"
• raw filter chips, row details' "Reviewed before commit"
• the schema tree's column types
• every results-grid column header (its tip is set in code, in CreateColumnHeader)

The pointer landed on the list item, card or header border behind each one. The schema tree's function, index, role and extension rows, NotifyMonitor's error lines, the conflict dialog's cut cells and the lock icon are the same shape; the fixtures don't show them, but the fix covers them by construction.

One mechanism, not a patch per site. Nimbus.Ui.Controls.ToolTipHitTesting, installed from App.Initialize, is a class handler. Any Panel, TextBlock, Border, ContentPresenter or templated control that gets a tooltip and has no background gets Brushes.Transparent as a current value:

• a background the markup or a style sets still wins
• one that goes back to null (a style that stops matching) is filled again
• disabled controls are left alone: Avalonia skips them in the hit test on purpose, and ToolTip.ShowOnDisabled is the opt-in

#332's two per-site Background="Transparent" setters are removed. TextBlock.statusMessage and CutTextTip move to nimbusUi, since nothing about them is Postgres-specific, and become DESIGN.md rule 21. CLAUDE.md records the evidence.

Paired PRs: Shman4ik/nimbusUi#6 (this subtree, plus the backlog nimbusUi had missed since 2026-09-28); Shman4ik/kubeNimbus#127 (the same handler there: 189 dead tooltips in 130 of 186 scenarios before, none after).

Verified

• dotnet build PgNimbus.slnx -t:Rebuild, and dotnet build src/PgNimbus.App -c Release -t:Rebuild: 0 warnings (trim/AOT analyzers on)
• PgNimbus.Core.Tests: not run; nothing in Core changed
• PgNimbus.App.Tests: 763 total, 749 passed, 14 skipped (the ones gated on a live database)
• TooltipReachTests goes red without the handler: 3 of 3 fail (the walk lists 203 dead elements)
• NativeAOT publish: not run
• Baselines: unchanged. Every scenario was rendered on main and on this branch, Windows, Release. 83 of 100 frames are byte-identical. The other 17 (security ×8, activity ×3, overview ×2, slow queries ×4) differ just as much between two renders of main, from their animated tab strips. The harness's own --baseline comparison passes.
• Published screenshots: nothing they show changed, so not re-rendered.

Anything left unverified: hovering in the real app on Windows/macOS by hand. The test hovers a history row with real pointer input on the headless platform, and the tooltip opens.

Checklist

• CLAUDE.md updated (the status-line paragraph under UI rule 6)
• Touches shared/nimbusUi: Take pgNimbus's copy (rules 16, 18–20), and tooltips that answer the pointer (rule 21) nimbusUi#6, and a paired kubeNimbus PR

🤖 Generated with Claude Code