add regex testing for non-alpha characters

[baha-bouali]

Fixes #4497

[a-zA-z] in the term-matching regex is a typo: the range runs from A (65)
to z (122), so it also covers the six characters that sit between the two
alphabets, which are [, \, ], ^, _ and `. And terms were matched across
them as if they were letters.

example

foo and the source string foo_bar text:

• before: \bfoo[a-zA-z]*\b matches foo_bar, highlighting the whole
  identifier as the term
• after: \bfoo[a-zA-Z]*\b does not match. In fact, _ is a word character, so there
  is no word boundary after foo

For the other five characters the old pattern didn't invent a match, it just
made an existing one too long: foo[bar] highlighted foo[bar instead of
foo.

The old suffix matching on actual letters is the same, so add-on still highlights
Add-Ons.

Tests

Added regression coverage for all non-alpha characters in Highlight.test.jsx,
I split them into two because the two groups behave differently:
_ is word character no mark at all, while the five non-word characters yield
exactly one mark equal to the term. Both cases fail against the old pattern.

make test-translate ran successfully after changes.