#27811 test(cmux NEXT remote desktop): the stale-retry test completes a real RFB handshake (rest of #26697)

Carries the rest of #26697. The session that opened #26697 can't be reached; #26697 stays open for its owner to close.

What it does now

#26557 (merged 2026-10-30) already landed #26697's first commit: RemoteVNCSession.reconnect(in:) cancels a pending automatic retry, and a first version of RemoteVNCReconnectTests.manualReconnectCancelsStaleAutomaticRetry. After a merge of feat-cmux-next, this PR is only #26697's second commit, a stronger version of that test:

• The fake transport completes a real RFB 3.8 handshake (None security, a one-pixel desktop), so the manual reconnect reaches .streaming instead of failing on the first read.
• The test waits for .streaming, keeps the connected client, waits past the 50 ms stale retry, and checks that there was still one attempt and that the same client is still streaming. The old test passed even when the connection had already failed.
• .timeLimit(.minutes(1)) and a defer { session.close() }, so a hang or a failed #require can't leak the session.

Observed

• At head 8a9032157b4b31ca1576815caf60a4534453fe15, all 5 RemoteViewVNCTests suites compiled and passed on a fleet Mac (fleet step cd6e79e3, about 11:21Z), RemoteVNCReconnectTests.manualReconnectCancelsStaleAutomaticRetry among them. This is the test-target compile evidence.
• Not run here: the fleet Release compile at this head. The first attempt returned "no result from the controller" while the compile controller was out of disk; it is being re-run.

Risk: low. One test file; no production code.

No capture: test-only change; nothing on screen changes.

Review

Independent review at the base-merged head: see the receipt comment. (The original head 147c4489df1f got a Codex SHIP with no findings.)

Agent session: agent session

🤖 Generated with Claude Code