Matches were previously found using `str.match(regexp)`, then
added as a `<mark>` at position `str.indexOf(matches[i])`.
This led to inaccurate highlights. For example:

- Text content: "Somewhere here"
- regexp search `/\bhe/g` (word boundary followed by 'he')
- `str.match(regexp)` correctly returns 'he', but not the position
  of the match
- `str.indexOf('he')` reports 5 (in the middle of 'Somewhere')
  instead of 10 ('here')

This is resolved by using `regexp.exec(str)`, which returns the index
of matched text, allowing word boundaries to function correctly (and
likely fixing some other devious but yet-undiscovered bugs).