Add shrinking to char arbitraries

[jmid]

This PR adds shrinking to char* arbitrary generators. In order to avoid shrinking a numeral char '7' into a non-numeral char it includes a dedicated shrinker. The printable case is not broken by the existing Shrink.char shrinker.

To eat my own medicine, I then proceeded to add Generator, Shrinker, and Statistics tests ...only to find bugs :open_mouth:

• it turns out the specification for QCheck.Gen.printable was wrong - it is nice to test it
• the distribution for QCheck2.Gen.printable revealed an off-by-one error in its internal range function, meaning that it would not produce, e.g., 'z'

[jmid]

I've refined the shrinker tests a bit to better exercise the expected behaviour. In doing so, I noticed that in the range '!'-'&' (character codes 32-38) QCheck2.printable shrinks towards '!' (lower character codes) contradicting the documentation of shrinking towards 'a' (97). I therefore updated the documentation to match the behaviour better. I admit it is underspecified - but fully specifying the behaviour becomes unreadable, IMO.

Also updated the CHANGELOG, so this should be good to go.

[jmid]

@vch9 Can you take a quick look at this since it finds and fixes an issue with the QCheck2.printable generator? :pray:

[vch9]

@vch9 Can you take a quick look at this since it finds and fixes an issue with the QCheck2.printable generator? pray

The fix seems correct to me

[jmid]

@vch9's comment made me realize that reusing the char shrinker for Gen.printable in QCheck(1) would break for '\n' (char code 10) by shrinking to a non-printable character. I've therefore committed an explicit Shrink.char_printable shrinker along with a few unit tests of the behaviour.

Looking at the code for Gen.printable in QCheck.ml I see the same approach of QCheck2 using a local array/table. This makes me think that the right solution(tm) would be to shrink towards the 'a' entry in this array (at index 97 - 32 or so). This could be done for both QCheck and QCheck2 and would clear up the weird QCheck2 character code jumping that is hard to specify.

[jmid]

I've now simplified the definition of QCheck2.Gen.printable. This makes the behaviour simpler to specify, as we always shrink towards 'a'. Bonuses:

• we don't need a list of printable_chars.
• this cuts 0.5 sec of the QCheck2 shrinker benchmark and reduces major GC runs slightly from 368 to 364.

I've also updated the tests and the CHANGELOG accordingly, so this should be good to go.

[jmid]

Rebased on latest master to resolve expect-test output conflicts