optimized list iterator some more

[omry]

Closes #613

Second optimization pass.
This is more impressive than the 7.7 -> 8.8x recorded improvement since there seems to be a regression compares to the measurements in #533.
Speed is still not great, but I this is as close as it gets given that we are probably losing a bunch of low level optimization done on native list because we need to unbox primitive items when we iterate.

Comparing test_list_iter[small_listconfig] by itself to what is in master shows about 50% speedup:

When it's pitched against the native list iteration it's "only" 130 times slower now, compares to 180 times slower before:

By the way: here is why this is important:
https://twitter.com/jackyliang42/status/1371530444529881094

[odelalleau]

Just one meaningful concern

(on the speedup from moving the in-line import => that's interesting, I've always wondered if there was a performance risk with these... good to know!)

[odelalleau]

Some questions

[omry]

@odelalleau I am merging this, let me know if you have any other feedback.