perf(parser): remove backtracking

[bgotink]

I noticed a performance regression when updating my benchmark to the new kdljs version.
Chevrotain is very slow at backtracking. Removing all backtracking results in a 7x faster parse time on my benchmark document and a 14x faster parse time on the KDL Playground's default document.

bench before x 1,695 ops/sec ±0.38% (99 runs sampled)
bench after x 12,230 ops/sec ±0.62% (96 runs sampled)

play before x 6,355 ops/sec ±0.33% (97 runs sampled)
play after x 88,774 ops/sec ±0.23% (99 runs sampled)

The document used in the benchmark

node \
{
	child; "child"; ###"child"###;
}

deeply { nested { node { but { like { really { deeply { nested { or { whatever; }; }; }; }; }; }; }; }; }

node \
  with="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
  #"no really,"#="a lot" \
	"of" "children" 4.20;

Playground document

foo 1 "two" three=(decimal)0xff {
  (thing)bar #true #false #null
}

This PR changes the way properties and arguments are parsed. The way this is defined in the spec requires unbounded lookahead due to the support for node-space between a property name and the equals sign, and unbounded lookahead only works in chevrotain if you use backtracking.
This PR changes the parser to use a different but equivalent ruleset to parse properties and arguments:

• Does it start with a tag? It is an argument
• Is it a non-string value? It is an argument
• Is it a string value? Check if it's followed by an equals sign.
  • If not, it is an argument
  • If yes, it is a property, check for an optional tag and a value

The propertyOrArgument rule returns whether it encountered trailing node-space, which is then used in the node rule.

[larsgw]

The benchmark file is now 10.680s instead of ~35s, thank you very much!