In my view the berachain testnet experienced unreliability due to
upstream issues in Comet.

* cometbft/cometbft#1940

While I cannot be sure, I think that some of the results that I got in
recent benchmarks of goleveldb actually represent clear bugs that could
under certain circumstances be used to halt chains.

* cometbft/cometbft-db#115

So here's the thinking:

1) drop goleveldb for pebble by literally specifying pebble statically
in the code
2) reduce max_bytes to 1mb
3) increase the size of the validator set and distribute VP evenly
4) torture the network, including torturing RPCs

Co-authored-by: Devon Bear <itsdevbear@berachain.com>