Port tests to Test2 and improve coverage

[jjatria]

This is an attempt at addressing the issues raised in #9. In particular, that to test random data it is more efficient to control the random seed than to iterate multiple times over the different paths.
As a baseline, this is what dzil cover reported before the changes in this branch:

All tests successful.                                                                                                
Files=12, Tests=30028, 16 wallclock secs ( 1.75 usr  0.23 sys + 15.38 cusr  0.27 csys = 17.63 CPU)                   
Result: PASS                                                                                                         
cover                                                                                                                
Reading database from /home/user/Data-Random/.build/0JJiU2UsDp/cover_db                                 
                                                                                                                     
                                                                                                                     
----------------------------------- ------ ------ ------ ------ ------ ------                                        
File                                  stmt   bran   cond    sub   time  total                                        
----------------------------------- ------ ------ ------ ------ ------ ------                                        
blib/lib/Data/Random.pm               94.7   77.5   36.2  100.0   99.8   76.1                                        
blib/lib/Data/Random/WordList.pm      97.7   58.3   36.3  100.0    0.1   82.6                                        
Total                                 95.3   75.4   36.2  100.0  100.0   77.1                                        
----------------------------------- ------ ------ ------ ------ ------ ------

And after

All tests successful.                                                                                                
Files=12, Tests=21,  4 wallclock secs ( 0.05 usr  0.01 sys +  3.92 cusr  0.20 csys =  4.18 CPU)                      
Result: PASS                                                                                                         
cover                                                                                                                
Reading database from /home/user/Data-Random/.build/LG4WKgKnS4/cover_db                                 
                                                                                                                     
                                                                                                                     
----------------------------------- ------ ------ ------ ------ ------ ------                                        
File                                  stmt   bran   cond    sub   time  total                                        
----------------------------------- ------ ------ ------ ------ ------ ------                                        
blib/lib/Data/Random.pm               99.4   89.8   57.8  100.0   86.7   86.7                                        
blib/lib/Data/Random/WordList.pm      97.7   58.3   45.4  100.0   13.2   84.0                                        
Total                                 99.1   86.3   56.6  100.0  100.0   86.2                                        
----------------------------------- ------ ------ ------ ------ ------ ------  

Although test coverage was high before, the suite took ~16 seconds to run on my machine. The new test suite takes a quarter of the time, and covers a little more code.

As an added bonus, the process of checking the different branches revealed a bug in rand_set() which meant that, calling it in scalar context, it would return an array of references rather than an array reference.

This changes use Test2::V0, but some of the tests still use Test::More. I'm not entirely sure how these tests should be ported, but this can be done separately, before the next release.