Cache all entries to calculate index

[delphinus]

I'm not sure that this is a good way...

The current build reads .index and calculates the ID in showing each entry. This diff enables to cache all entries formerly and search them pretty fast.

This effects most significantly, it improves 49s → 10s to nb list on my box.

nb listresult

> time ./nb list > /dev/null

________________________________________________________
Executed in   49.56 secs   fish           external
   usr time   20.24 secs  149.00 micros   20.24 secs
   sys time   27.77 secs  814.00 micros   27.77 secs


> time ./nb list > /dev/null

________________________________________________________
Executed in   10.92 secs   fish           external
   usr time    6.19 secs  132.00 micros    6.19 secs
   sys time    4.91 secs  884.00 micros    4.91 secs

[delphinus]

Hmm... nb needs to support bash <4.2 that has no associative arrays? Then, we should think a compatible way to work with it.