Complete the public-alpha hardware and network acceptance pass

[FelineStateMachine]

Close the remaining gap between protocol or kernel evidence and actual couch gameplay.

Acceptance criteria:

• Physically headless cold boot creates Gamescope and starts Sigil before the first SSH login.
• A physical controller attached to Portal controls an actual game, including neutral release on disconnect.
• Left and right mouse buttons are consumed by Gamescope or the target application, not only observed in evtest.
• A sustained gameplay soak records cadence and hitch percentiles, queue drops, A/V skew, CPU, GPU, and memory without latency growth.
• A difficult-NAT relayed session is identified in diagnostics and measured separately from a direct session.
• Exact release-candidate binaries repeat the reconnect and second-client rejection gates.

[FelineStateMachine]

completed in testing, but serves as UAT

[FelineStateMachine]

Hardware UAT is now executed rather than deferred. Exact-main workflow run 29919920109 built commit 7920c5d214344f859e0cd5987433a3b303eb85f3; PR #36 preserves the fail-closed runner and sanitized evidence.

GPD Pocket 4 / Bazzite 43 results: fixed 1280x800 sustained 58.701 fps with zero post-encode drops; native 2560x1600 completed at 29.759 fps with zero drops. Forty authenticated Iroh/MoQ + grouped-v3 sessions completed with zero sequence gaps, all input acknowledgements and forced-IDR recoveries passed, invitation replay was rejected, and the original service/config/identity/release were restored and independently verified.

This closes the local direct-path hardware/reconnect portion. Difficult NAT/relay, slow-decoder pressure, physical-controller gameplay, and headless cold-boot acceptance remain separate acceptance legs.