refactor($parse): separate tokenizing vs parsing more

[jbedard]

Fixes a.b and a .b generating different getterFns (which was mentioned a while back in #8901 (comment))
Fixes #9893 ({'': ...} being converted to {"''": ...})
Fixes {.: ...} parsing as {'.': ...} instead of throwing
Fixes #9131 (similar to how #9229 fixes it)

Also simplifies (imo) a few areas in parse.js. For example...

• tokens no longer have any fn (which had 2 meanings before)
• token text is always the token text (previously it was the raw value for strings/numbers)
• .consume() is used in a few more places instead of .expect() and will output a "Unexpected end of expression" instead of .expect() returning false and failing later
• the tokenizer no longer tries to figure out if a ".identifier" is a method invocation or not, that's now in the parser and much shorter

[lgalfaso]

why is (this.index < length-1) needed?

[jbedard]

Because isNumber(false) is true when there is nothing after the '.'. Could also add a is-string check to isNumber which might be better then this length check?

[lgalfaso]

well, false is not a number, so isNumber should be fixed :)

[lgalfaso]

otherwise, lgtm

[jbedard]

Updated.

I also changed one more expect || throw to consume (https://github.com/angular/angular.js/pull/9431/files#diff-780de070ede30180f6aedb6c5f7d57caL614).

[btford]

looks good, provided there are no perf implications.

[jeffbcross]

Moving this to backlog to review/merge after 1.3.0 GA since the diff is so substantial.

[jbedard]

I've rebased this...

[pkozlowski-opensource]

This PR also fixes #9893

[IgorMinar]

@petebacondarwin we should get this in in 1.3.4 or 1.3.3 before it gets bit rotten.

@lgalfaso or @rodyhaddad could help with the review.

[lgalfaso]

I'll take care of the review

[lgalfaso]

landed with fbad280