Improve LMDB cardinality estimation

[hmottestad]

Improve LMDB cardinality estimation with a file-parsing / page-walk implementation that reads triples/data.mdb directly.

[kenwenzel]

We can also try to copy the estimation implemented in libmdx.

[hmottestad]

Yeah. I read a bit about their implementation. It's what inspired me.

With my implementation the estimates are very accurate, which unfortunately creates a bunch of regressions for queries that now have a worse query plan :(