say: canonical-data needs improvement.

In the canonical data for the "say" problem:
https://github.com/exercism/x-common/blob/master/exercises/say/canonical-data.json

* The descriptions should be descriptive.

In most cases the description is the same as the output.
```
{
"description": "zero",
"input": 0,
"expected": "zero"
},
```

The description should describe the test, and why it is useful to test that particular number.

* Should test all available numbers.

A purely test driven implementation of these tests would miss out a lot of numbers.
(particularly the teens and tens)