Refactor utf-8 strings and invalid strings in test code

[tompng]

This pull request resolves the problem mentioned in the source code comment in input_keys.

def input_keys(input, convert = true)
   # Reline does not support convert-meta, but test data includes \M-char. It should be converted to ESC+char.
   # Note that mixing unicode chars and \M-char is not recommended. "\M-C\M-\C-A" is a single unicode character.
   ...
end

"\M-[char]" should be removed from test code because it is an invalid byte sequence string.
Reline only accepts "\e[char]" which is sent by Meta-[char] key input in modern terminal emulators.

After invalid byte sequence are removed, we don't need convert option in method input_keys because valid strings does not need to skip convert.

Other changes

def convert_str(input) can be simpler.
If the input needs unicode_normalize before conversion, then the prerequisites of the test (input unicode-unnormalized text) are broken. We just need to omit the test if the test encoding is not utf-8.