Well, let's start with that N mod 255
doesn't convert number to 8bit byte, but a byte without highest possible value, 255. You were looking for either N mod 256
or N and 255
(somewhat known assembly trick to do some divides faster).
Now, if your source is cryptographically sound, then any amount of bits cut from it will be sound (except might collide faster). However this isn't case when cutting a value that isn't power of 2! In your example, let's assume we are cutting 8-bit values (0-255) to your N mod 255
(0-254). In this case we will get 1-254 with same probability, but we are twice as likely to get 0, and we will never get 255. This obviously makes skew in your data, which is usually combated by cutting last values that would cause such skew (in my example, we would reject 255 and draw value again).
Of course all above holds in your specific case of ISAAC.