SegmentMK: Optimize reading of large binaries

Currently the SegmentMK splits all binaries to 4kB blocks, which adds some overhead when reading large binaries. When writing we already combine up to 64 such blocks to a 256kB bulk segment that is then written in a single operation.

We could use the same idea also when reading: if a sequence of two or more blocks are located next to each other in the same segment, we can merge those blocks into one big block for reading a larger chunk of data at a time.