Garbage Collection S3 Mark Improvements with `-m` flag Resource Usage

In testing garbage collection with a large repository gitlab-com/gl-infra/infrastructure#9033, passing the -m flag causes a higher than expected use of system resources: greater than 14 Gigabytes of memory. This is seen even at 100 requests per second, which is lower than current default of 350. In comparison, without this flag and with 500 requests per second, memory usage topped out at around 2.5G. This is relatively heavy usage, but not extreme considering the higher workload.

The Mark stage with the S3 walk improvements has not been as extensively test with the -m flag and these tests reveal that there could be stability issues when running garbage collect against larger registries.

We are nearing the time to release a new version of the container registry, so I believe that we should not use parallel walk in s3 until we have had the opportunity to more fully test garbage collection using this option, with the intention to release in the next milestone.