fix(setup): verify WSL before Local AI downloads

[RomneyDa]

Summary

• move the existing WSL install-and-verify stage ahead of Local AI runtime and model acquisition
• fail before the 23 GB model download when WSL cannot start because hardware virtualization or Virtual Machine Platform is unavailable
• keep the GPU compatibility check first, and align onboarding progress rows with the new execution order

This is a one-commit follow-up to openclaw#1178.

Validation

• git diff --check - passed
• .agents/skills/autoreview/scripts/autoreview --mode local - passed, no accepted/actionable findings
• focused dotnet test - blocked locally because dotnet is unavailable on the macOS controller
• required ./build.ps1, shared tests, and tray tests - blocked locally because no native Windows/Crabbox host is configured

Real behavior proof

Not verified / blocked locally: this path requires a Windows host with WSL absent or hardware virtualization disabled. The pipeline ordering tests assert that EnsureWslPlatformStep precedes both Local AI download steps, and the progress-page contract asserts the same visible phase order.

Rubber-duck review

The reviewer found that the progress UI still grouped early WSL verification with later WSL networking. The implementation now splits those phases and removes the stale “Verify Local AI before WSL setup” claim.