Overhaul io gate structure

[st0012]

I think the current IO gate design has a few non-functional issues:

1. Their source don't indicate that they're IO gate classes. Usually for task-specific components like this, the files would be contained in the same folder and their would be a abstract super class to define shared methods/states.
2. These IO gate classes are not used as classes but singleton modules. This means
   • States are all defined as class variables right when the file is required.
   • If we change any state on Reline::IOGate, like in tests, we need to reset it back through redefining the methods, which is pretty hacky.

This PR addresses the issues by:

• Moving io gate classes into a dedicated folder
• Converting io gate classes into normal Ruby classes and change class variable states into instance variables
• Renaming GeneralIO to Dumb to better reflect its role

[st0012]

The readline test failures are caused by readline-ext manually setting IOGate. I think making it call Reline.test_mode directly may address the issue.

I'm not sure what causes the windows failure though and I don't have a windows machine. Any help would be appreciated 🙏

@tompng Do you want me to include the changes in #659 here?

[tompng]

I think this @original_screen_size and IOGate.set_screen_size(*@original_screen_size) is not needed because we don't modify @original_iogate anymore.

(perhaps this is causing windows test failure)

[tompng]

Looks good 👍

For #659, I think separate pull request is better

[tompng]

I think default value of encoding should be nil, or we need to check RUBY_PLATFORM.
If @encoding is always set, we can't get into elsif RUBY_PLATFORM =~ /mswin|mingw/ branch in def encoding and this seems to be the reason CI (irb) is failing. (but I can't reproduce the failing in my local environment...)

[tompng]

CI / readline can be passed by setting Reline::GeneralIO = Dumb.new, perhaps as deprecated constant