Improve performance tuning slice capacity

[jsign]

I worked on top of #17.

• deleted unused function: Just to cleanup agreed on @derekparker review.
• created a count of leaves per node: this allows to create the keys slice in collect with the correct capacity to avoid allocations while appending.
• in collect the nodes capacity is set to a lower-bound which save more allocations.

Here's the benchmark comparison to master now:

benchmark                   old ns/op     new ns/op     delta
BenchmarkTieKeys-8          5098          2385          -53.22%
BenchmarkPrefixSearch-8     393791        118249        -69.97%
BenchmarkFuzzySearch-8      6001483       4151116       -30.83%

benchmark                   old allocs     new allocs     delta
BenchmarkTieKeys-8          56             3              -94.64%
BenchmarkPrefixSearch-8     5548           3              -99.95%
BenchmarkFuzzySearch-8      47089          8009           -82.99%

benchmark                   old bytes     new bytes     delta
BenchmarkTieKeys-8          840           256           -69.52%
BenchmarkPrefixSearch-8     71628         13120         -81.68%
BenchmarkFuzzySearch-8      793796        476874        -39.92%