Shifted option for variable collation elements

[lucafavatella]

Depends on #18 (included in this PR). Supersedes #15 #16 #17.

Notes:

1. I am a bit concerned that the conformance tests doc says

   Compare that string with the string on the previous line, according to the UCA implementation, with strength = identical level (using S3.10).

   but tests succeed without having implemented the identical level at all... so the conformance tests currently do not really need the identical level.

2. What is more our tests calling the conformance tests keep strength to default, that is tertiary and not quaternary. And conformance tests pass - even the shifted ones. But based on common settings combinations doc

   “Ignore punctuation” (completely): strength=tertiary alternate=shifted

   that suggests that the conformance tests for shifted do not test the punctuation... or the code actually handles the shifted ternary as quaternary (i.e. keeping L4).

3. I am pretty confident that some other language / language library already has Unicode tests further to the UCA conformance tests by Unicode. We should find and import such extra tests.