Improve the gas estimation algorithm

[rach-id]

          > Before we use Z score, IMO we should verify that the gas prices in historical blocks form a normal distribution.

Having a normal distribution is optional in our case IMO since we don't need spot on values. There are three cases:

• a flat distribution: where the values will generally be close the to the mean which is fine
• a U distribution (I mean just the shape of the distribution forming an U or a W or similar): where we have two cases: either the low prices will be higher than 10%. Or the high prices will be lower than 90%. but I guess that's fine.

I tried to convey in the ADR that median of top 10% is likely easier and good enough so we don't need to use z scores.

got you 👍 I avoided implementing the percentile because it means implementing a bit of complicated math method and maintaining it and I wanted to avoid that. If we can find a library we can use, that has an open source license, I'll definitely change this implementation.

Originally posted by @rach-id in #4257 (comment)

[rach-id]

also relevant: #4238 (comment)