Consider making --no-rocksdb-compaction the default

[mvines]

As observed in #10199, the rocksdb compaction phase of Blockstore::purge_slots() on a fully loaded ledger can take 30-200 minutes. Is this even worth it?

• If no, replace --no-rocksdb-compaction with --rocksdb-compaction
• If yes, consider breaking out compaction from Blockstore::purge_slots() into a separate process that runs less frequently than every 512 roots (which currently effectively means it's always running, but only after the --limit-ledger-size shred limit is reached). Perhaps at the start of every epoch.

[mvines]

I have two API machines running side by size with a 1TB ledger, one with --no-rocksdb-compaction and one without.

• --no-rocksdb-compaction present: DISK READ: 1079.34 K/s | DISK WRITE: 2.50 M/s
• --no-rocksdb-compaction absent: DISK READ: 87.06 M/s | DISK WRITE: 89.11 M/s

[mvines]

After observing some nodes for a week, I feel that we do want to compact still by default but we don't need to compact that frequently. Once a day ought to be ample, which is what #10414 implements