Closes #97027

What Problem This Solves

Fixes an issue where managed browser automation on a Linux service host without DISPLAY or WAYLAND_DISPLAY either stopped on an opaque launch failure or forced callers to infer recovery from human text. The same deterministic preflight failure could also consume the managed-launch cooldown and block a subsequent explicit headless retry.

Why This Change Was Made

Implicit managed launches now use the existing Linux headless fallback, but an explicit headed choice remains authoritative and fails before Chrome starts. That failure keeps the existing actionable message and carries a closed no_display_for_headed_profile reason plus typed details through local routes, the HTTP client, Gateway RPC, permission routes, and opt-in browser-node proxy envelopes. Mixed-version nodes retain the legacy error contract unless the caller advertises the new envelope.

User Impact

Server-side browser automation can start headlessly when the user has not chosen a mode. Explicit headed configurations are never silently overridden; agents and programmatic callers can branch on a stable reason, then recover with openclaw browser start --headless, OPENCLAW_BROWSER_HEADLESS=1, a headless profile/config choice, or a real display server. Deterministic no-display preflight failures no longer poison launch cooldown.

Evidence

• Blacksmith Testbox: 146 focused browser tests passed, followed by the complete changed-surface gate (extension production/test type checks, extension lint, import-cycle and repository guards): https://github.com/openclaw/openclaw/actions/runs/29024582684
• AWS Crabbox jade-crayfish (cbx_8f76bd8fabee): exact patch-content digest 49bd9941171726a79ee407cd7f99f640d7495425ebbe12f301b100cc6f64baf2 verified; build run run_d281f5376b04 passed; live run run_2cb350ea7889 used real Playwright Chromium 149 to load “Example Domain” through both implicit linux-display-fallback and explicit one-shot --headless recovery.
• The same real Dev Gateway returned the structured explicit-headed failure with zero Chrome processes before/after rejection.
• Live run run_2cb350ea7889 also required a real Gateway-backed openai/gpt-5.4 agent turn to discover and call the browser tool; its JSONL transcript was checked for the exact browser start call, the causal failing tool result, actionable no-display text, and no embedded Gateway fallback.
• Fresh autoreview: no accepted/actionable findings after repairing a discovered node-proxy stale-target 404 regression.