Rename AsciiWordChar and optimize Codepoint

[pitdicker]

This renames AsciiWordChar to Alphanumeric.
And just after making the Avoid low bits PR, I changed AsciiWordChar to use the low bits (?!). Better use a shift than a mask now.

Adding the [inline] attribute to RangeInt::new and RangeInt::new_inclusive helps LLVM to calculate the range and zone at compile time if the range is constant. That makes it a better choice than the current bitmask or even simple_range for Codepoint.

Benchmarks before:

test distr_uniform_ascii_char ... bench:       2,785 ns/iter (+/- 59) = 1436 MB/s
test distr_uniform_codepoint  ... bench:       9,792 ns/iter (+/- 36) = 408 MB/s

After

test distr_uniform_alphanumeric ... bench:       2,654 ns/iter (+/- 54) = 1507 MB/s
test distr_uniform_codepoint    ... bench:       2,799 ns/iter (+/- 5) = 1429 MB/s