Improve lexing of ternaries that include symbols in Ruby lexer

[pyrmont]

Ruby's rules for how it parses ternaries are complicated. This is all the more the case if the ternary contains symbols. The current lexer uses a simple test to determine whether a colon demarcates the branches of the ternary: is the colon immediately followed by another colon. While this rule suffices for many cases, it can have problems with symbols.

This PR replaces the simple rule with two rules:

1. Simple case: The simple case is where there is whitespace surrounding the colon. If this is the case, the colon demarcates the branches of the ternary.

2. Complex case: The complex case tests whether there are no more colons or backslashes on the line. If there aren't any, the colon should be treated as demarcating the branches (see this StackOverflow question for edge case examples).

This fixes #1038.

[pyrmont]

@jneen It doesn't look to me like this causes any regressions in the visual sample (although I take your point in #1038 that this is an exceptionally tricky area to lex correctly). Do you have any thoughts?

[jneen]

I haven't taken a super close look at this, but definitely check the year on that stackoverflow post - Ruby's syntax has changed a great deal, and you don't need backslashes for code like this:

condition
  ? antecedent
  : alternative

[jneen]

...also a?b:c is completely valid. ><

[jneen]

Hang on I take that back.

a?b :c # parsed as a? (b) (:c), syntax error for missing comma
a?b:c # parsed as a?(b: c), b is a symbol
a ?b:c # parsed as (a) ? (b) : (c), a proper ternary operator
a ?b :c # same as above

[pyrmont]

@jneen Perhaps I'm now optimising for the tests but the samples are now all lexed correctly.

This isn't counting the lexing of methods ending in ?. Those aren't matched by the ternary rules but are subtly broken, I think. That seems like something that should be solved in a different PR.