Implement rolling trace log

[benbjohnson]

This pull request changes the "trace log" so that it can be written to disk and the logs will automatically be rolled. Trace logging can produce A LOT of data quickly but we typically only care about very recent logs when an issue occurs.

Usage

The existing -tracing flag still works and can be used to enable the trace log via STDOUT. There is also a new tracing config section for enabling the trace log printing to disk:

# The tracing section enables a rolling, on-disk tracing log. This records every
# operation to the database so it can be verbose and it can degrade performance.
# This is for debugging only and should not be enabled in production.
tracing:
  # Output path on disk.
  path: "/path/to/trace.log"

  # Maximum size of a single trace log before rolling. Specified in megabytes.
  max-size: 64

  # Maximum number of trace logs to retain.
  max-count: 10

  # If true, historical logs will be compressed using gzip.
  compress: true

This default configuration will produce a file called /path/to/trace.log and once that log reaches 64MB then it will be rolled over to a new file with a time-based name (e.g. /path/to/trace.2022-12-15T14-11-07.991.log.gz). Once 10 logs files have been produced, the oldest one will be deleted when a new log file is written.