This change adds a guard to a flag to ensure that the `eval` will produce a value. Without this change, a flag like `my-flag` will go through the `eval` as `$my-flag`. Since bash doesn't permit dashes in variable names, the resulting value will be `-flag`. Because this is a non-zero length, the first iteration through the for loop on `${e}` will set in `FLAGS`, then break out. This leads to the improper result of FLAGS having `-flag` instead of the desired value of `my-flag`.

By testing that flag will produce a value before trying `eval`, we can be sure that the first loop iteration will proceed and allow the processing of the second iteration on `${flag}` (which will correctly capture `my-flag` in our example).

${!flag+x} is used for indirect variable expansion. If flag contains the name of a variable, ${!flag} will expand to the value of that variable. The +x part ensures that the expression evaluates to true if the variable exists, even if it's empty.

This fixes codecov#175.