Cloud Instance Launch Visibility and Diagnostics

Goal: every Your Cloud launch (AWS, GCP, Azure; Ubuntu and Windows) reaches online and runs a task, or shows why within minutes. Never a silent vanish, an open-ended "Launching", or a billed VM left behind. - Fix the causes found on dev: AWS Ubuntu instance created but never reports in; GCP Windows (n2-standard-2) never dials in (suspects: 900 s tailnet key vs first-boot sysprep, GCE startup-script runner, egress to Tailscale/Git downloads); GCP Windows order vanished from Nodes/Hosts with no audit entry. - Boot reason on every cloud: `cloud_boot_watch` reads GCP serial port and Azure boot diagnostics as it does EC2; roster and node page show `infra.boot_report.reason` (row turns stalled at once) beside "Stalled". - Operator-only tail of the scrubbed boot log on the host page. - State label after the instance exists says what it waits for (not "Launching"; not "?"/"None" on Windows rows). - A newly created node in the current org opens without "Hidden by the org filter". Out of scope: Jaah Hosted pricing; lifecycle actions (own task); self-hosted nodes. Done means: on dev, a fresh AWS Ubuntu and a GCP Windows node each reach online and run a task (screenshots); a launch that cannot report in shows a reason within 15 min (test); unit tests pin each stall cause.

Boot watch reads AWS/GCP/Azure boot logs; cloud tailnet keys last 2 h; stalled reasons and operator boot log in UI; failed launches stay listed and audited. #6641+#6648 merged, gate green. Dev: AWS Ubuntu online; GCP Windows boot-loops after WSL reboot (draft #9082); task run unverified.