Performance improvements

[veelenga]

Two main improvements:

1. Cache parsed AST in a Source. So Parser will parse content only once.
2. Use one unified visitor with multiple callbacks instead of multiple visitors to traverse AST.

This improves performance significantly, for example running it on a Crystal repository now takes ~1 second, instead of 6 seconds that it was before (1179 files).

Benchmarks:

== Compare:
  1 source  12.69k (  78.8µs) (± 1.12%)       fastest
 3 sources   6.21k (161.15µs) (± 1.34%)  2.04× slower
 5 sources   1.87k (533.92µs) (± 1.24%)  6.78× slower
10 sources 869.12  (  1.15ms) (± 2.20%) 14.60× slower
20 sources  487.1  (  2.05ms) (± 1.12%) 26.05× slower
30 sources 361.01  (  2.77ms) (± 1.06%) 35.15× slower
40 sources 361.28  (  2.77ms) (± 0.86%) 35.12× slower
== Measure:
  0.010000   0.020000   0.030000 (  0.018274)

in comparison to master:

== Compare:
  1 source    1.2k (832.79µs) (± 2.44%)       fastest
 3 sources 697.94  (  1.43ms) (± 2.08%)  1.72× slower
 5 sources 204.45  (  4.89ms) (± 5.59%)  5.87× slower
10 sources  103.1  (   9.7ms) (± 3.42%) 11.65× slower
20 sources  61.21  ( 16.34ms) (± 1.62%) 19.62× slower
30 sources  47.33  ( 21.13ms) (± 1.53%) 25.37× slower
40 sources  47.56  ( 21.03ms) (± 1.16%) 25.25× slower
== Measure:
  0.060000   0.040000   0.100000 (  0.073773)