LPD: remove gain bits for proper raw preview

In writing tests for !90 (merged), I just remembered that the raw preview is currently just computed from the raw input buffer. This is convenient. As we know, the gain of each pixel is encoded in the uint16 after the 12 bits of image data. This is smart.

Unfortunately, those two in combination make raw preview meaningless for non-zero gain values. A proper raw preview should just be the raw image data - one could have that open next to the gain preview to get the full effect in a reasonable-looking way.

It's not hard to add (we already write corrected data and extracted gain values to separate outputs during "correction"; just write the raw data, too, immediately after doing & 0x0fff to a separate output), it's just outside the scope of !90 (merged) (which is already huge), so making this issue with the intend to fix after.