enable cache_db auto compaction

[mononaut]

the shared rocksdb store constructor disables auto compaction to speed up initial indexing:

electrs/src/new_index/db.rs

Line 312 in 5d39070

db_opts.set_disable_auto_compactions(true); // for initial bulk load

and then we have to explicitly re-enable it on each store instance later:

electrs/src/new_index/schema.rs

Lines 363 to 365 in 5d39070

	start_fetcher(self.from, &daemon, to_add)?
	.map(|blocks| self.add(&blocks, Operation::AddBlocks));
	self.start_auto_compactions(&self.store.txstore_db);

electrs/src/new_index/schema.rs

Lines 373 to 375 in 5d39070

	start_fetcher(self.from, &daemon, to_index)?
	.map(|blocks| self.index(&blocks, Operation::AddBlocks));
	self.start_auto_compactions(&self.store.history_db);

but when the cache_db was added in [2699560], apparently auto compaction was never re-enabled for that store.

this means

• the rocksdb L0 SST files for the cache_db grow unbounded every time we flush the cache_db memtable to disk, as we do after every precache run.
• point lookups for scripthashes not in the cache_db (i.e. most /address/:address queries) get more and more expensive, since they have to search every L0 SST file to confirm there are no hits.
  • lookups for scripthashes that are in the cache are still fast, since a) they're usually still in the memtable, and b) if not, the latest entry is probably in one the first SST files we check.

in normal operation this isn't very noticeable because the cache_db is small and rarely flushed, but if you did something silly like reboot electrs and re-run the precache every hour then read performance for less popular addresses could eventually become severely degraded.