- The only reasonable means of mirroring the data layout.

  The '{}' delimiters mark the extent of the binary writes.

  The primitives 'label' and 'scalar' are directly supported and correspond
  to known byte widths.

  Using "List<scalar>" was a bad choice, since this triggers unpleasant
  tokenizing behaviour. Instead use 'scalars' as a provisional placeholder
  to indicates a list of scalar values. However, there is currently no
  support for actually handling lists of scalars, for several reasons:

   * The information is not available at compile-time.
     The cloud or parcel must be queried. And it must be guaranteed
     that this value is consistent for the entire cloud.

   * Binary output of lists is currently not great for determining the
     the encoded width:
        - A zero-size list is a single '0'.
        - The leading size is a non-constant number of digits.
        - There are prefix/suffix newlines need to be tagged and
          skipped.

  The special '*' (glob) token indicates that the remaining content
  has a dynamic variable length and nothing reasonable can be known
  about their sizes. This is exemplified by the collision records.