Increase size of default leveldb cache to improve random read performance

[barton2526]

This continues a discussion from Slack yesterday. Our current dbcache parameter defaults to 25 MB, which has not been changed since we forked off from Blackcoin in 2014. Bitcoin had a leveldb cache size of 100 MB in 2014, then bumped it to 300 MB in 2016 (morcos/bitcoin@32cab91) and finally to 450 MB in 2017 (bitcoin/bitcoin#10133).

I am in favor of a more conservative size, and I think 100 MB would be a good starting point to see what performance benefits we can expect from this change. Leveldb's internal benchmarks (http://www.lmdb.tech/bench/microbench/benchmark.html) suggest greatly improved read speeds with a larger cache (1.29x for sequential and 1.47x for random). If we can benchmark ourselves to get real world numbers for the change that would be even better.

As a reminder, the dbcache parameter can be easily changed by the user, so memory constrained devices can continue to use 25 MB (or even lower) if they do not want this change.

[barton2526]

Related: checking the code, we currently have a TODO to implement the -reindex parameter. We could perform more accurate testing of this change (reindexing the block files with no network variables) with -reindex.