Refactor Reline::Unicode ed_ vi_ em_ split_by_width methods

[tompng]

Depend on #759 (split some commits to it)

Refactor Reline::Unicode vi_ ed_ em_ methods

Rewrite them with -250 lines.
Update questionable test case.

Performance

# before (slow because time complexity was O(n^2))
irb(main):002> Reline::Unicode.em_backward_word('a'*2000,2000)
processing time: 0.265469s
=> [2000, 2000]

# after
irb(main):002> Reline::Unicode.em_backward_word('a'*2000,2000) 
processing time: 0.001357s
=> 2000

[st0012]

Does it make sense to encode gcs all at once before looping through them?
Also, all tests still pass even if I removed these encoding calls. Can we add a few cases for them?

[tompng]

encode will change bytesize, so I have to fix all encode(Encoding::UTF_8).grapheme_clusters in this file to return the correct bytesize in original encoding. I'll also add a test.