Improve logging if sandbox startup fails

[aaaaalbert]

When developing sandboxes that the nodemanager can control, it happens that things go wrong, and that spawning the sandbox results in an error. The current nmAPI.py doesn't handle this gracefully -- errors are simply ignored, and so a sandbox startup error results in

1, a mildly opaque error message in seash that the sandbox Did not start in a timely manner, but
2, no sign whatsover in the nodemanager log (which this issue proposes to fix).

Note that if you happen to start the sandbox directly from the command line (as opposed to through the nodemanager, i.e. usually seash), you will get to see any sandbox startup errors. Alas, you don't always get to launch a sandbox like this (think Android, see SeattleTestbed/utf#67), and even if you do, the startup code in nmAPI could still contain bugs.

---

So there! Let's make nmAPI collect any sandbox error output and send it to the nodemanager log