Fix parsing ambiguity around adjacent > and = tokens

[cirras]

This PR fixes a grammar ambiguity around adjacent > and = tokens. (Fixes #318)

We now create >= tokens during parsing by combining the >/= tokens, instead of combining them at the lexing stage.
(For consistently, we're doing this for "less than equal" tokens as well)

This allows us to only create >= tokens when we're parsing a binary expression.

As a result, we can now successfully parse code that looks like this:

const Foo: TArray<Byte>=[1, 2, 3];

Previously this would create a >= token in the middle and make everything fall over.

[fourls]

Looks good! A couple questions.

[fourls]

Looks good!