GatedStream.Dispose does not end pending reads, so net472 readiness rows wait out the 5 s release bound #441

tests/OpenCode.Sdk.Tests/Support/GatedStream does not override Dispose, so closing it does not end a pending ReadChunkAsync. A real pipe's reader ends when the pipe closes.

On net472, StreamReader.ReadLineAsync() takes no cancellation token. A gated stdout therefore waits out PipeOutputPump's 5-second release bound instead of ending at the close. sdk-runtime measured the two recorded-exit rows in OpenCodeServerReadinessTests at master: 5.07 s and 5.05 s on net472, against about 1 s for the rows that use real pipes.

The rows pass, so this is a realism and wall-time gap in the fixture, not a red test. The cost is a fixture that differs from production in exactly the path those rows test (stream end at close), plus about 4 s per row on net472.

Suggested fix: GatedStream.Dispose faults or ends the pending reads, the way a closed pipe does. The two rows would then finish at the close, near the real-pipe timing.

Out of TST-E4's scope (its roadmap steps don't name it). The readiness rows are in RT-E2 C's in-flight files, so the fix goes after RT-E2 C merges. Reported by sdk-runtime during TST-E4 planning, 2026-10-17.