[xandreeddev]

Closing notification history can let a passive desktop widget reclaim keyboard focus before a cold browser receives keys. This change separates keyboard capability, explicit user demand and native ownership. The framework performs one-shot acquisition and verifies retention after grab teardown; caret/focus paint follows actual keyboard activity. Domain ports remain independent of native objects and secure clients opt out.

The original ba5 cold-browser pointer and Tab/Enter routes pass, including trusted keyboard input and all nine cleanup checks per case. This PR remains a draft: fast inline-widget acquisition, outside handoff and reacquisition still need a valid complete native trial. The earlier held-button/completed-click failures retain their exact outcomes; their guest delivery order was not measured.

Validation on source ba5b8a51:

• Arch40433468095 passes 17 stages, 880 targeted tests, 94 named regressions and a genuine 121-input/480-image software-Vulkan catalogue. Coherent six-client/root-greeter deployment preserves settings/security.
• Both original canonical cold URI cases pass three feature checks, one trusted unmodified nonrepeat KeyA and nine cleanup checks each.
• Native lock/PAM V3 passes six checks, including new compositor acknowledgement, exact native locker, authenticated unlock, retained same-invocation clean exit and wallpaper resume. Earlier failed attempts remain preserved.

The separately compiled/deployed d77 control-flush candidate passes 17 stages/885 tests/99 named regressions and fresh export. Its original cold URI pointer and Tab/Enter cases also pass all three feature/nine cleanup checks each. A completed Notes click delivers the first key and normal undo. The later closed V10 diagnostic establishes a test-transport problem: the guest receives the outside KeyA 4.332 ms before the tablet press, and the pointer never reaches the browser. This does not establish a Lucent focus failure or infer earlier trials' guest order. The observer closes both descriptors; eight cleanup checks pass but original-cursor restoration fails, with no later recovery claimed. Correct physical pointer preparation and retesting remain pending. d77 source/captures are not promoted into this ba5 PR.

Still missing: valid fast first-key/outside/reacquisition acceptance; secure pixels/login and suspend regression; eligible normal memory at or below 300 MiB; source-rate/60 FPS and smooth interaction measurements; physical-device and broader application coverage. Source-specific reports and historical flags remain unchanged. No full-goal or performance acceptance is claimed.