Added test case for non-interleaved MCU crash

In cases of, eg. a 2x2 Y channel and two 1x1 chroma channels, a
progressive pass on the Y channel would previously consume blocks
at a rate of 4 blocks per MCU, instead of the correct 1 block per
MCU. In this example image, once 8 blocks had been consumed from the
stream, the decoder thought it was on just starting the third MCU,
instead of the correct ninth MCU (starting on the second row).

Thus, it thinks there are still 6 remaining MCUs before the next
restart marker, and it tries to continue reading from the stream.
Given that the rest of the byte are all 1 bits, this fails as the
huffman tree consists of just a single 0b entry.

Image was generated in GIMP and exported with the following settings:
* Optimize
* Use restart markers, interval 1 MCU row
* Progressive
* Subsampling 4:2:0 (chroma quartered)

Author: quilan1