Consider branching factor of 32 or 64

[sgraf812]

In my very POC Array-mapped trie implementation, benchmarks suggested that a branching factor of 32 or even 64 would result in better performance on my PC. The paper on RRB trees for example mentions that 32 is a sensible balanced choice. 64 favors lookup performance over insert performance.

[sjakobi]

Sounds worth playing with to me.

Regarding the implementation, is this really as simple as changing

unordered-containers/Data/HashMap/Internal.hs

Lines 2241 to 2242 in 405859e

	bitsPerSubkey :: Int
	bitsPerSubkey = 4

to 5 or 6!?

I ran the test suites with these settings and they surprisingly passed!

[Boarders]

@sjakobi I think that is the case. Did you measure how it affects performance at all?

[sjakobi]

Did you measure how it affects performance at all?

Not yet. Help with this would be very welcome! :)

[sjakobi]

For reference: #317 includes changing the branching factor to 32.