Support route-specific HTTP failures in network fixtures

Drive 2.0.1 can apply connection-level chaos, but reproducing an OpenCode TUI location-sync regression required a small loopback proxy to return HTTP 500 for only GET /api/agent while GET /api/location and SSE remain healthy. The same setup needs to heal that route and click Retry without restarting the instance.

A scoped HTTP fault override in the network fixture would make this repeatable without a Bun preload/proxy. Expected: select a method/path, return a deterministic status/body, inspect matching request counts, and remove the override. Current workaround: preload a loopback proxy for the script-owned TUI only, forward other requests, and toggle the selected response from the isolated fixture. OpenCode base tested: d9c85d8d95. This is a Drive fixture capability request, not the OpenCode bug itself.

Start by locating the network fixture and its isolated controls, then compare them with the loopback proxy workaround described in the issue. Done means a selected method/path can return a deterministic status and body, matching request counts can be inspected, the override can be removed, and the route can heal for a Retry without restarting the instance.