Add agent selector to Skills page

In multi-agent OpenClaw setups, the top-level Skills page requested skills.status without an agentId and only showed the default agent workspace. Skills stored under other configured agent workspaces were functional but invisible from the main Skills UI.

Added an agent selector to the Skills page, threaded selected agentId through skill status, Skill Card, security verdict, install, and ClawHub update flows, and reset stale reports, messages, verdicts, and unsaved API-key drafts when switching agents.

• Makes the Skills page usable for multi-agent OpenClaw installations without forcing users through the Agents page or manual workspace inspection.
• Preserves agent workspace isolation by routing installs and updates to the selected configured agent instead of always targeting the default workspace.
• Reduces stale-state risk by preventing skill reports, verdicts, messages, and edit drafts from leaking across agent switches.

• Added controller and view tests for selected-agent skills.status, security verdicts, Skill Card, install payloads, and selector rendering.
• Added gateway tests proving explicit-agent ClawHub installs route to the selected agent workspace.
• PR validation included real gateway selected-agent workspace proof, 107 focused tests, scoped formatting/lint checks, diff checks, and clean final autoreview.