Configure the Thanos Compactor to downsample metrics and manage retention in object storage

Steps

1. Run `thanos compact --data-dir <dir> --objstore.config-file=bucket.yml` as a single singleton instance against a given bucket/block-stream — only one Compactor may safely operate on the same stream (grouped by external labels) at a time.
2. Downsampling runs automatically in two passes: 5-minute-resolution blocks are produced from raw blocks older than 40 hours, and 1-hour-resolution blocks are produced from 5m blocks older than 10 days; disable it entirely with `--downsampling.disable` if not needed.
3. Set `--retention.resolution-raw`, `--retention.resolution-5m`, and `--retention.resolution-1h` to durations (e.g. `--retention.resolution-raw=30d`) to control how long each resolution tier is kept in the bucket; leaving them at the default `0d` retains that resolution forever.
4. Keep retention at each resolution greater than or equal to the minimum age required for the next downsampling pass (e.g. raw/5m retention should exceed 10 days) so data isn't deleted before it can be downsampled into the next tier.
5. Run the Compactor continuously with `--wait --wait-interval=5m`, or invoke it as a periodic batch/cron job, since by default it exits once it finishes a compaction pass.
6. Tune `--compact.concurrency`, `--consistency-delay` (default 30m, delay before newly-uploaded blocks are considered for compaction), and `--delete-delay` (default 48h, grace period before deletion-marked blocks are physically removed) to match bucket size and object-storage consistency guarantees.