rebalance lazy compression levels

[Cyan4973]

This is a proposed update to rebalance compression levels.
It makes levels stronger than then used to be in v1.5.0, and closer to (though generally still faster than) v1.4.9, which achieves a more gradual scale towards level 13 (the first btlazy2 level).

It's still difficult to create a "good" level 12, because rowhash is just very fast, so there's only so much that can be slowed down by increasing parameters to the max.

note: having looked at 256K table, there is no obvious update, the current table already tries to be balanced, and there is still a large gap between strongest lazy2 and fastest btlazy2, simply because lazy2 is very fast, even when pushed to its maximum.

[Cyan4973]

arf, tons of updates to do in the regression test ... :(

[Cyan4973]

After extensive testings on x64, x86 and aarch64, this new map of compression levels produces a more "regular" speed / compression ratio curve on these systems (than the map currently defined in dev).

[felixhandte]

Discussed my concerns about memory monotonicity offline. Looks good to me.

This is definitely more regular: