Remove io_optimized / sbq_speedup storage mode (#195)

The `io_optimized` / `SbqSpeedup` storage mode follows the DiskANN
approach more closely than `storage_optimized` / `SbqCompression`, but
by storing neighbor vectors in `SbqNode`, the index size becomes bloated
and benchmarks performed poorly. Thus it was never documented externally
or used anywhere afaik. This CR simplifies the code by removing this
storage mode.