[sylvesterkaczmarek]

Problem

HTTP VM address lookup requests use the fixed 30-second client timeout even when VMs().IP requests a longer server wait. A requested 120-second wait therefore reaches the transport with a 30-second deadline. The WebSocket path already extends its request timeout to the requested wait plus the usual HTTP allowance.

Change

Share that existing wait-aware client selection with ordinary HTTP requests, and rename the internal helper to reflect both callers. Preserve longer explicitly configured timeouts, disabled client timeouts, and earlier caller-context deadlines. Requests without a wait keep their existing behavior; the shared client is not mutated.

Validation

• Broad suite excluding VM/runtime integration packages: 124 tests and 101 subtests passed.
• The client package passes with the race detector. Changed-code lint reports zero issues; git diff --check passes.
• New tests exercise the public IP method and inspect the effective transport context deadline without sleeping through a long timeout. Two cases fail on unchanged production code; controls cover zero wait, longer/disabled client timeouts, and an earlier caller deadline.
• The existing WebSocket timeout test remains covered.

Go 1.27.1 on macOS arm64. No real controller, VM, or prolonged network wait was used. No public API, authentication, query-format, or dependency changes.

Go documents how the client-wide timeout covers the complete request in http.Client.

Reproduced against f101ade5d8fdcad5e9c346472698649e7107a715.