Updated Kernel conversion methods

[sampersand]

This is a merging of the previous individual PRs for conversion methods into one, larger PR. (#1433 , #1434, #1435, #1436, #1437, #1438, #1439). Look at them for a little more in-depth details, adn some discussion topics. Summary of changes is below:

• Kernel#Array: Updated to represent the true definition: Only .to_ary and .to_a are used for conversions, and everything else is boxed up in an array.
• Kernel#Complex: Added the .to_c variant.
• Kernel#Float: Changed from Numeric | String to just be _ToF, which is what's used `internally.
• Kernel#Hash: The argument is no longer an Object: Instead, .to_hash is required for its argument, with the special case of nil and empty arrays (and only empty arrays) returning empty hashes.
• Kernel#Integer: Updated to reflect the ability to pass implicit .to_str and .to_int to all variants, and only .to_i (not being a subclass of Numeric) for the one-argument case
• Kernel#Rational: Oh boy, this method is a mess. Check out the original issue for more details. tl;dr: If the first argument isn't _ToInt | _ToR, but is Numeric, it attempts to do division with it, which allows for returning arbitrary types.
• Kernel#String: Simply cleaned up naming; it was semantically correct beforehand.

For all conversion methods with an exception: bool field, two variants have been added: (..., ?exception: true) -> type and (..., exception: bool) -> type?. This reflects the fact that with exception: false (or an unknown boolean value, which may be false), the return type is nil.

Additionally, all the exception: bool variants, when given invalid types but with exception: false, will return nil. That's also been accounted for.

[soutaro]

Looks good. Some newline problems to make the CI pass.

[soutaro]

I will take a look at the CI fails.