A failure to write trace logs can cause processes to run out of memory

A [recent issue](https://github.com/apple/foundationdb/issues/5405) revealed a case where an fdbserver process that couldn't write its trace files was running out of memory. It would be better to cap the amount of memory we keep for trace logs and start dropping them if we exceed it.

Start by reading the linked issue #5405 and tracing how fdbserver writes its trace files. Identify where unwritten trace logs are retained, then verify that retention is capped and excess logs are dropped so a failed trace-file write cannot exhaust process memory.

Written by the indexing model from the issue text.