Refactor Reline::Unicode ed_ vi_ em_ methods

[tompng]

Depend on #759 (split some commits to it)

Refactor Reline::Unicode vi_ ed_ em_ methods

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

Description of refactor

Using get_prev_mbchar_size and get_next_mbchar_size in a while loop is extreamly inefficient.
Using take_while to an array of grapheme_cluster is simple and fast enough.

Performance

Scenario: paste 2KB text aaaaa...aaaaa to IRB and press OPTION+LEFT.

# 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.