fix: detect machine update failures

[rianmcguirefly]

Change Summary

What and Why:

Previously, when we gave up on an update and reverted a machine to its previous version (eg. if pulling the image from the registry failed), flyctl saw the machine in the "started" state (the old version still running) and treated the deploy as successful. Health checks would pass against the old image, so users had no indication their new code wasn't deployed.

Affects both flyctl deploy (in-place updates) and standalone flyctl machine update.

How:

Check the machine version is what we expected after wait succeeds.

Related to:

---

Documentation

• Fresh Produce
• In superfly/docs, or asked for help from docs team
• n/a

[rianmcguirefly]

=== RUN   TestFlyMachineRun_port
    test_env.go:105: workdir /tmp/TestFlyMachineRun_port1156148536/001
    result.go:59: expected successful zero exit code, got 1, for command: /home/runner/work/flyctl/flyctl/bin/flyctl machine update -a preflight-gha-25306166751-1-2026-05-de7ev3qz 48e2479ae49718 -y --port 80/tcp:http --port 1001/udp [stdout]: Updating machine 48e2479ae49718
Error: machine 48e2479ae49718 update failed: machine remains on previous version 01KQRY9AA1G8862ADXZH7GRPPK (state: replaced)
         [stderr]: Error: machine 48e2479ae49718 update failed: machine remains on previous version 01KQRY9AA1G8862ADXZH7GRPPK (state: replaced)
--- FAIL: TestFlyMachineRun_port (20.19s)

Well that's not supposed to happen. Investigating...