Length of Scan segment can be off if there are restart markers

[mauricefisher64]

In reader.rs read_scan_data function should kick out not just for 0x00 but also for any restart segment (RSTn).

             if byte != 0x00 {
                self.current_marker = Some(byte);
                break;
            } else {
                for _ in 0..ff_count {
                    data.push(0xFF);
                }
                data.push(byte);
            }

I think it should be something like this:

fn in_entropy(marker: u8) -> bool { matches!(marker, 0xd0..=0xd7 | 0x00) }

              if !in_entropy(byte) {
                self.current_marker = Some(byte);
                break;
            } else {
                for _ in 0..ff_count {
                    data.push(0xFF);
                }
                data.push(byte);
            }

[mauricefisher64]

I think there also needs to be a rollback of the position pointer - 2 when !in_entropy(byte) is true since you have read the 0xFF and byte (the next tag). Otherwise the size will be off by two bytes since this code just detected the next tag.

[vstroebel]

Even if not technically correct, the current implementation treats restart markers like normal segments because I need this information for my personal use case. Your solution would completely remove the existence of restart markers from the dump.

One possible alternative is to extend the Scan struct to include all restarts as individual data parts. But this could make reading/dumping broken JPEG file more complicated.