Parentheses in a regular expression capture whatever is matched by their content. This means that the part of the string that was matched is accessible in one of the special variables $1, $2, $3. So in your case, $1 simply means the content of the <% %> pattern. For more information you can read perlretut.

For the eval part, the documentation states:

EXPR is parsed and executed as if it were a little Perl program
[...]
This form is typically used to delay parsing and subsequent execution of the text of EXPR until run time.

So the aim of the eval is to turn the content of the <% %> pattern into a "little perl program", and obtain its last value.

There is a catch though, there is not one, but two evals in your code because a replacement operation is of the form s/REGEX/STRING/. So the thing on the right side is a string (simple text), not code to be executed. Among the three options on the replacement operator (g, e and s), the e means "execute", or simply eval.

So in a nutshell: this operation $inp =~ s/<%([^=].*?)%>/eval $1/ges takes whatever is inside a <% %> pattern (except if the content starts with an equal sign), and executes it as a little perl program, before replacing the pattern by the last value in that program. In your example, the last value is $test.