G4: Expose bounded session metadata in session list

[SarthakWade]

Problem

session list currently returns session names and isolation flags only. Agents cannot identify which session is on the relevant page or how old a session is without inspecting each one and changing agent-control state.

Split from #54 G4.

Contract

• Add an architecture decision defining metadata fields, freshness, privacy, bounds, and engine parity.
• Extend typed session details with current HTTP or HTTPS URL, page title, creation time or age, isolation mode, and lifecycle availability.
• Collect metadata without enabling agent control, taking a semantic snapshot, activating a window, or changing page state.
• Redact URL userinfo and bound URL and title lengths. Mark page-derived title and URL data as untrusted.
• Return per-session metadata failures as bounded status fields so one broken or navigating session does not fail the whole list.
• Keep deterministic name ordering and preserve the existing sessions compatibility array.
• Do not expose cookies, storage, credential aliases, authentication state, native window identifiers, process internals, or filesystem paths.

Acceptance criteria

• Normal, isolated, navigating, closed-race, and failed-session paths are covered.
• Both WebKit and Chromium return the same metadata shape.
• Long or hostile titles and URLs are bounded and safely marked untrusted.
• Listing sessions has no focus, navigation, agent-control, or authentication side effects.
• Protocol schema, generated SDK contracts, CLI docs, unit tests, and macOS/Linux E2E coverage pass.
• Existing callers using only sessions and isolated remain compatible.