provide tools to better figure the LRU cache size

[colinsurprenant]

current default size is 1000. provide information to help on finding the best cache size. this information should probably be exposed through the eventual metrics features.

this relates to the upcoming #12

[colinsurprenant]

As I discussed in both
#12 (comment)
#12 (comment)

I tried to express that I think there is a relatively "small" upper bound in the number of user agent combinations in a given timeframe (release of a new browser version will create new combinations) and I think this upper bound can probably estimated using any relatively popular public facing web site web access logs.

out of curiosity I checked the Wikipedia user agents datasets. These dataset are a sample so it will not contain all user agent combinations but nonetheless I think it should provide a feel for what order of number of unique user agents we can expect to see generally speaking.

This is a simple wc -l of the 4 files provided:

  250 desktop-editors-parsed.tsv
   50 desktop-readers-parsed.tsv
   47 mobile-editors-parsed.tsv
   15 mobile-readers-parsed.tsv
  362 total

As we can see, we are not talking about potential thousandS unique user agents here. Any specific use-case will certainly end-up having to deal with less combinations.

So, I think that our 1000 default LRU cache size is probably big enough to hold all user agent combinations. The LRU nature of the cache will work with us to evict the rate freak user agent combinations and I believe we can probably expect a near %100 cache hit ratio.