Improve Unicode support with grapheme clusters

[jquast]

• I read the contributing guidelines
• and that's why I include some flake8 and typing/mypy fixes
• I also included comprehensive tests
• many were designed with TDD (started failing, succeed after change).

Issues fixed by this PR

• I did not discover any open issues, I am surprised!
• You must not have many CJK and emoji uses.

What does this implement/fix?

Problem: asciimatics text utilities wrongly split up grapheme clusters (emoji ZWJ sequences like 👨‍👩‍👧, regional flags like 🇨🇦, skin tone modifiers, combining characters, etc) -- this causes display corruption in the SpeechBubble, or incorrect width calculations and padding in other places due to missing grapheme support.

Solution: Integrate with wcwidth>=0.5.0 by using:

• wcwidth.iter_graphemes() for iteration in _enforce_width_ext(), _find_min_start(), _get_offset().
• wcwidth.wrap() for grapheme-aware word wrapping in _split_text().
  • Note that you can also pass-through **kwargs for all of the additional built-in features supported by wcwidth.wrap() also, like requested by Add a way to let the user provide a custom text wrapping function #386
• wcwidth.ljust() for line padding in SpeechBubble

Any other comments?

• I notice that there is a choice to "ignore unicode" for performance improvement, but I can suggest that wcwidth has many "fast path" checks for pure-ascii strings to return len(string) and so on, along with lru_cache, the performance is negligble to always support unicode since, related downstream automatic benchmarking results can be viewed here of upgrade of wcwidth:

View benchmarks of wcwidth 0.2.14 to 0.5.0:

	Benchmark	BASE	HEAD	Efficiency
⚡	test_center_ascii	104.8 µs	62.1 µs	+68.87%
⚡	test_rjust_cjk	760.6 µs	640.2 µs	+18.81%
⚡	test_center_cjk	763.5 µs	642.6 µs	+18.81%
⚡	test_truncate_ascii	33,519.5 µs	83 µs	×400
⚡	test_center_ansi	678.1 µs	359.2 µs	+88.78%
⚡	test_truncate_cjk	14.2 ms	6.8 ms	×2.1
⚡	test_length_cjk	749.7 µs	633 µs	+18.45%
⚡	test_truncate_emoji_zwj	5.2 ms	1.6 ms	×3.4
⚡	test_rjust_ansi	673.7 µs	356.1 µs	+89.19%
⚡	test_truncate_ansi	46.1 ms	23.2 ms	+99.18%
⚡	test_length_ansi	672.2 µs	488.9 µs	+37.5%
⚡	test_length_ascii	99.2 µs	60.4 µs	+64.23%
⚡	test_rjust_ascii	99.8 µs	58.9 µs	+69.33%
⚡	test_ljust_ansi	670.5 µs	355.3 µs	+88.74%
⚡	test_ljust_ascii	100.2 µs	58.9 µs	+70.1%
⚡	test_ljust_cjk	759.6 µs	640.7 µs	+18.55%
⚡	test_length_emoji_vs16	739.7 µs	670.7 µs	+10.28%

• I can also suggest it is possible to use try/except of import wcwidth, with hasattr, making it an optional co-dependency, to help solve for Reducing number of dependencies #95 and Non-unicode version? Wcwidth is a resource hog. #155

Let me know if you would like any such changes or additional PR's, happy to help.