Performance improvement 3: simplify queries

[msm-code]

I'm a bit disasppointed in this PR - TBH I expected another big win, while it turned out to be mediocre.

Nevertheless there are some improvements here, and the code is not too complex, so I think it's worth merging (I rejected a few ideas that turned out to be complex to implement for almost zero benefit).

As usual, results can be seen in this repo: https://github.com/msm-code/ursa-bench temporarily hosted here: http://65.21.130.153:8000/hdd_all.html

Ok, but what's going on in the code?

Basically we flatten queries:

AND(AND("A", "B"), AND("C", "D")) becomes AND("A", "B", "C", "D").
OR(OR("A", "B"), OR("C", "D")) becomes OR("A", "B", "C", "D").

And we handle minof special cases:

1 of ("A", "B") becomes OR("A", "B")
2 of ("A", "B") becomes AND("A", "B")

(should be merged after improvements 2)

[msm-code]

After my adventures in #197, I've decided to check performance of this PR again. Again, let's use time because we don't have a proper counter in place (#198).

This PR with warm cache:

(venv) root@mquery ~/ursa-bench # time python3 ursabench.py rules/signature-base/yara/* > /dev/null
real    1m18.954s
user    0m13.420s
sys     0m0.700s

This PR after dropping page caches:

real    70m7.672s
user    0m17.272s
sys     0m0.832s

This is a bit faster than #197, but the difference is minimal. In the current form it doesn't bring enough benefits to merge, I think.

On the other hand, maybe we could prioritise ngrams depending on their use count. But I'm not yet ready to implement this.