Improve performance of reference definition list parsing

[RomanHotsiy]

Fixes #996

Looks like the reference definitions cannot interrupted by any of the elements if I read the spec correctly.

None of the tests fail.

By removing the redundant termination logic checks the speed of parsing long list of reference definitions improves dramatically.

~2.5x on the benchmark included in this repo, but in our example where we have 1000 definitions it's almost 20x!

benchmark before

./benchmark/benchmark.mjs block-ref-list
Selected samples: (1 of 28)
 > block-ref-list

Sample: block-ref-list.md (782 bytes)
 > commonmark-reference x 29,907 ops/sec ±0.24% (96 runs sampled)
 > current x 6,429 ops/sec ±0.23% (98 runs sampled)
 > current-commonmark x 7,261 ops/sec ±0.13% (98 runs sampled)
 > markdown-it-2.2.1-commonmark x 36,093 ops/sec ±0.20% (94 runs sampled)
 > marked x 55,087 ops/sec ±2.06% (99 runs sampled)

benchmark after

./benchmark/benchmark.mjs block-ref-list
Selected samples: (1 of 28)
 > block-ref-list

Sample: block-ref-list.md (782 bytes)
 > commonmark-reference x 29,025 ops/sec ±0.50% (94 runs sampled)
 > current x 15,550 ops/sec ±0.35% (94 runs sampled)
 > current-commonmark x 16,762 ops/sec ±0.63% (95 runs sampled)
 > markdown-it-2.2.1-commonmark x 35,562 ops/sec ±0.23% (91 runs sampled)
 > marked x 55,408 ops/sec ±0.18% (98 runs sampled)

[RomanHotsiy]

oh, it looks like it can be improved more by removing the redundant code in the reference parsing.

I'll adjust it later today.

[rlidwka]

Here's a test which would have different behavior:

[foo]: /url 'title
 - - -
'

[foo]

It was <hr>, now it's a reference link.

[RomanHotsiy]

Superseded by #1004