gtk.cfg: Remove pure annotation from g_str_has_prefix/suffix

[correctmost]

g_str_has_prefix and g_str_has_suffix are not technically pure because they can log to the console if a precondition fails.

This partially reverts commit 7cc7c0b.

---

Some notes:

The original gtk.cfg change hasn't been released yet, so the revert shouldn't cause any churn for users.

GLib has code like this:

gboolean (g_str_has_prefix) (const gchar *str,
                             const gchar *prefix)
{
  g_return_val_if_fail (str != NULL, FALSE);
  g_return_val_if_fail (prefix != NULL, FALSE);

  return strncmp (str, prefix, strlen (prefix)) == 0;
}

Most real-world code is treating this function as safe to call inside of an assert, even though it could technically have side effects if one of the preconditions fails.

I removed the pure annotation to err on the side of correctness and pedantry, even though most projects would view the warnings as "false positives" from a practical perspective.

Here is example usage from QEMU:

char *qemu_chr_get_filename(Chardev *chr)
{
    ChardevClass *cc = CHARDEV_GET_CLASS(chr);
    const char *typename;

    if (cc->chr_get_filename) {
        return cc->chr_get_filename(chr);
    }

    typename = object_get_typename(OBJECT(chr));
    assert(g_str_has_prefix(typename, "chardev-"));
    return g_strdup(typename + 8);
}

[chrchr-github]

Maybe it would be better to keep this test case and suppress the warning to make the desired behavior more explicit.

[correctmost]

Good idea, that also allowed me to add a comment for future readers.