Handle missing items in get_fs() and tspa_mx()

[marklhc]

The standard errors are not constant with missing data. Currently lavaan gives a different covariance for different missing patterns. Need to incorporate this into get_fs() for row-specific errors and loadings.

Then show an example of analyzing this in tspa_mx().

• Compute factor scores with missing data
• Compute standard errors with missing data
• Include scoring matrix, fsL, fsT, etc, for each missing pattern
• Unit test showing that our function gives same results as lavaan
• Provide an example with OpenMx using individual-specific standard errors.

[marklhc]

May need to restructure the internal structure for compute_fspars()

[marklhc]

Add attributes to augment_lav_predict() for easier specification of cross-loadings and error covariances