Improve performance of HTML parser on JVM

[tindzk]

The HTML parser incurs a significant slowdown as the nesting level
increases:

$ bloop run pine-bench-jvm -- slow
[...]
Benchmark: Parse HTML w/o attributes
- depth=2:
  units: 7
  iterations: 591733
  run time: 3384 μs/it ± 7
- depth=6:
  units: 127
  iterations: 11610
  run time: 171586 μs/it ± 504
- depth=10:
  units: 2047
  iterations: 56
  run time: 36148809 μs/it ± 74820
- depth=14:
  units: 32767
  iterations: 1
  run time: 9353666666 μs/it ± 174704194

Summary:
  Unit growth: 18.1x, 16.1x, 16.0x
  Run time growth: 50.7x, 210.7x, 258.8x

This slowdown can be attributed to the rest() function in Reader.
It calls data.drop() which on the JVM creates a copy of the string
rather than pointing to the same memory.

Scala.js' drop() implementation has the expected semantics such that
the run time performance is roughly linear to the number of nodes in the
tree:

$ bloop run pine-bench-js -- slow
[...]
Benchmark: Parse HTML w/o attributes
- depth=2:
  units: 7
  iterations: 92592
  run time: 21624 μs/it ± 128
- depth=6:
  units: 127
  iterations: 5229
  run time: 382531 μs/it ± 182
- depth=10:
  units: 2047
  iterations: 312
  run time: 6479611 μs/it ± 134542
- depth=14:
  units: 32767
  iterations: 17
  run time: 119013071 μs/it ± 1455066

Summary:
  Unit growth: 18.1x, 16.1x, 16.0x
  Run time growth: 17.7x, 16.9x, 18.4x

After applying the optimisations, the parser will behave similarly on
the JVM:

$ bloop run pine-bench-jvm -- slow
[...]
Benchmark: Parse HTML w/o attributes
- depth=2:
  units: 7
  iterations: 991955
  run time: 2048 μs/it ± 51
- depth=6:
  units: 127
  iterations: 45471
  run time: 43403 μs/it ± 412
- depth=10:
  units: 2047
  iterations: 2523
  run time: 777550 μs/it ± 10750
- depth=14:
  units: 32767
  iterations: 147
  run time: 13759510 μs/it ± 160258

Summary:
  Unit growth: 18.1x, 16.1x, 16.0x
  Run time growth: 21.2x, 17.9x, 17.7x