fix htmlinputstream, tokenizer to use codepoints and build strings as late as possible

[DartBot]

<img src="https://avatars.githubusercontent.com/u/5479?v=3" align="left" width="96" height="96"hspace="10"> Issue by sethladd
Originally opened as dart-lang/sdk#16704

---

There's a todo explaining, but I think we can radically simplify this class and base it on codepoints instead of building up string chunks. I'd also like to remove the RegExp that is used for "charsUntil"

This could be done in two phases also, first simplify inputstream then tokenizer

[DartBot]

<img src="https://avatars.githubusercontent.com/u/5479?v=3" align="left" width="48" height="48"hspace="10"> Comment by sethladd

---

Was https://github.com/dart-lang/html5lib-now-on-dart.googlecode.com/issues/23

[DartBot]

<img src="https://avatars.githubusercontent.com/u/17034?v=3" align="left" width="48" height="48"hspace="10"> Comment by kevmoo

---

Added Pkg-Html5Lib label.

[DartBot]

<img src="https://avatars.githubusercontent.com/u/17034?v=3" align="left" width="48" height="48"hspace="10"> Comment by kevmoo

---

Removed Library-Html5lib label.

[jmesserly]

funny to see bugs bouncing around :) ... this was originally filed by me, but I don't think it's super high pri anymore. The only perf issue we've seen is the string concat, which is (more or less) fixed now.

[jmesserly]

also is covered by #1094 :)