The old algorithm collapsed null bytes early on.  The new algorithm
keeps the null bytes until the 32-byte chunks have been read.

Parsing hooke/test/data/vclamp_mfp3d/Line0004Point0001.ibw with the
old algorithm gave labels as

  [[], ['RawDeflLVDT'], [], []]

The new algorithm gives

  [[], ['', 'Raw', 'Defl', 'LVDT'], [], []]