Improve support for emojis+cursor w/grapheme clustering

[jquast]

Problem

Copy and paste this test sequence to call prompt_toolkit.prompt('Give me some input: '):

 👨‍👩‍👧 👩‍❤‍👨 👩‍💻👋🏿 ❤️⭐ 🇯🇵🇩🇪 café niño Åộ 中文!.

Moving the cursor over and around emojis get strange. insertions become chaotic. Cursor position becomes indeterminate (even negative!), input result becomes more corrupted with user confusion as draws become corrupted. This is briefly described in #274 by @jonathanslenders:

Notice that it still requires multiple cursor movements (left/right arrow) to move across these characters.

wcwidth-integration-before-trimmed.mp4

Solution

Close #274 "Handle decomposed unicode characters" (2018) through careful integration of new functions, wcwidth.iter_graphemes and wcwidth.grapheme_boundary_before.

wcwidth-integration-after.mp4

Details

• Out of respect for the size and popular use of this library, I took the care to rewrite "pty test helper" code I first created for blessed and later in telnetlib3, to provide a real tty to a subprocess for true integration testing. maybe i will publish it as a pytest plugin someday.
• It covers only the essentials but you might enjoy this new test accessory to provide better coverage elsewhere, I have simulated special keypresses, like HOME, END, etc, hope it helps.

[jquast]

This might show some improved test coverage for this reason

feel free to extend this test to get those previously unreachable lines

[jquast]

A more advance version of this test could help verify screen draw reactions and effects of resize on SIGWINCH signal, for example.

[jquast]

These tests are simplified but provide pty coverage, the highest level possible exercise, which is nice because it is also necessary to test these changes.

More improvement could be done to also verify screen draw effects of the REPL, these only verify output after sending return key, where repl integration print()'s result.