virtualization-mcp v1.6.0

Sandbox reliability release. The naked-PC install harness went from
fleet-wide timeouts to green end-to-end runs with screenshots.

Naked-test harness (opt-in E2E proven on email-mcp: pass, 12/12 pages)

Singleton lifecycle: 60 s teardown wait (incl. the vmmem Hyper-V worker),
240 s boot verification via job-dir markers, one retry, loud
failed_step: "harness" RESULT instead of ghosts (src/.../sandbox_management.py,
webapp/backend/app/main.py)
vmmemWindowsSandbox.exe tracked in running-check and terminate
Prefer [remote "origin"] URL over submodule file:// URLs; full
fleet-standard git path for the local --bare mirror
Headless discipline: only declared -NoBrowser/-Headless switches are
passed; Edge opens by exe path (no association dialog); stdio stays attached
New opt-in e2e_click stage: full-stack startup, generic UIA sidebar walk
(assets/sandbox/Invoke-E2EClick.py, no OCR), E2E.json + e2e-*.png
start.bat forwards args; root start.ps1 bootstraps uv via
Require-Command (NAKED_PC_INSTALL_STANDARD)

FastMCP Streamable HTTP mounted at /mcp/ (with lifespan) — kills the 404
Dashboard REST bridge: GET /mcp/tools, POST /mcp/tools/call,
POST /api/mcp/tool (Tools Console + helpers work again)
13 portmanteau tools verified live through the bridge

Fail-soft VBox init: import no longer crashes where VBoxManage is absent;
VM ops return a friendly error when called (proven in a VBox-less Sandbox)
PyInstaller spec: joserfc + _datetime hiddenimports (FastMCP 3.4.5)

start.bat serves the MCP server, not the :10701 web dashboard — matrix
health verdicts for this repo need an entry-point decision (see BUILD_LOG)
Uninstaller leaves %LOCALAPPDATA%\Virtualization MCP logs behind (155 KB)
WebView bridge-OK text pending dashboard-side check (CUA smoke 10/11)