Handle INFINITY as literal

Currently, it's impossible to use Float::Infinity as a literal in signatures:

 def initialize: (Integer | Float::Infinity bytesize) -> void
#=> RuntimeError: Neutered Exception RBS::Test::Tester::TypeError: TypeError: [MyClass#initialize] ArgumentTypeError: expected `::Integer | Float::Infinity` (bytesize) but given `Inf
inity`
 def initialize: (Integer | Infinity bytesize) -> void
#=> RuntimeError: Neutered Exception RBS::Test::Tester::TypeError: TypeError: [MyClass#initialize] ArgumentTypeError: expected `::Integer |Infinity` (bytesize) but given `Inf
inity`
 def initialize: (Integer | Float::INFINITY bytesize) -> void
#=> TypeError: class or module required

Consider the case of a buffer with bounded or unbounded size, common in network software. In such a case, you want to tag the buffer size precisely (as an integer), or leave it open-ended (with Float::INFINITY). It's unacceptable that, for instance, 3.4 is set as buffer size; however, as per the current constraint, the only way to set a signature that would be tolerant to infinity would be:

 def initialize: (Integer | Float bytesize) -> void

which would tolerate other floats.

I don't know whether there are other variables handled specially, but I'd argue that this Infinity should be handled the same way as 4in any given rbs signature.

No file or test is named in the issue. Reproduce the Float::Infinity and Infinity signature examples, then trace the parser and type-checking entry points that handle literal types. Done means a signature accepts infinity as a distinct literal without accepting arbitrary Float values.