The regex is wrong, it will also match attr names like `a-` or `-b?` or
even `-`, which are of course not valid.
Since the whole concept of sacy is centered around parsing
developer-supplied html and doesn't use a full-blown parser, I feel
the compromise is acceptable here in order to keep the regex simple.
This puts the html data attributes under the key `data` in the work unit
For example, for the following script tag:
<script data-some-option="blah" data-other-options="blih" ...
The work unit array will contain:
'data' => array(
'some-option' => 'blah',
'other-option' => 'blih'