We now use nested data types to explicitly describe the matrix structure
and to avoid using magic numbers to select specific columns.

The top-level fields are:

* weight: the particle weights;
* prev_ix: the parent indices;
* state_vec: the particle state vectors; and
* lookup: sample indices for lookup tables (if required).

A further aim is to allow models to specify the state vector data type,
including the ability to have named fields and non-scalar entries.

In general, simulation models should only have access to state vectors,
while observation models and summary tables should have access to the
full matrix.