create `pred_obs_mc` method

[murphyk]

• create pred_obs_mc method for all the estimators that samples from bel(theta), using the full cov or DLR approximation, and then returns a SNC matrix of predictions on the N test points. Similar to xmap_yhat_sample and xmap_yhat_sample_lofi_dlr, but methods of the estimator class, not buried inside a notebook.

[murphyk]

For the 1d regression demos,

• change xmap_yhat_sample_lofi_lin in demos/nonstat-1d-regression to use the predict_obs_cov method.
• For the linear plugin approximation , plot +- confidence intervals on each predicted output, rather than drawing independent samples from the marginals p(y_t|xt; mu{t|t-1})..

[gerdm]

See

• bf332de3662eb9620912f6e74049990025c54680
• d02c71c0f6a3ed6c1b46d8bc1e1050551c8b7f8c
• 7b0458f08ea37d7b76d58993cfa03ba1418116d5