Improved trace thread error management

There are currently a few problems with handling trace logging thread errors that need to be addressed, as discussed in this [forum post](https://forums.foundationdb.org/t/processes-oom-fdbmonitor-doesnt-restart/1855/4):

1. The failure to flush a trace file shouldn't block the shutdown of a process.
2. We should decide what the intended behavior of a process with a failed trace thread is. Currently, it continues running but eventually runs out of memory, which doesn't seem like the correct choice. I think we should either kill it quickly or let it run indefinitely.
3. Record somewhere whatever details we can about the error that caused the trace thread to die.
4. https://github.com/apple/foundationdb/issues/1977

Start by reading the linked forum post and issue #1977, then locate the existing trace-thread shutdown, flush, and error-handling paths in the repository. Done means the failure behavior is explicitly settled and implemented for shutdown, process lifetime, and recording available error details.

Written by the indexing model from the issue text.