Add a new binary to the server build, named mongor. This will be the workload replay client.

The implementation should be a simple stub for later tickets to build upon.

The bazel changes and rough outline of main can largely be brough over from the spike implementation.

The exact nature of the parameters Atlas will provide is TBD (e.g., will the client handle fetching from S3), but for now we will assume a single argument, which is a path to a directory (which will contain one or more recording files).

The actual replay implementation will be added in a separate library which the binary can depend upon; this was useful during the spike for testing.