Investigate better backtracking support.

[bd82]

Backtracking is evil and inefficient, but sometimes one may not care as much about performance. So perhaps it is possible to improve the backtracking support.

Things to consider:

• Better Performance for backtracking.
  • Reuse backtracking final results instead of re-parsing.
  • Reuse backtracking partial results when attempt to backtrack multiple alternations?
• Ability to control CST building when backtracking (on/off) toggle.

[bd82]

may become redundant with the introduction of adaptive predict (LL(*))