Tutorial covariances changed to squares

[oharrald-Dstl]

Not an essential change, but it may provide some clarity. Perhaps when defining covariance matrices / noise in the tutorials, we could use squares (eg. noise_covar = np.diag([2**2]) instead of np.diag([2])) so that it is clear that the user defines variances rather than standard deviations in model instantiations.

[jmbarr]

And what would you do for off-diagonal elements?

[oharrald-Dstl]

I suppose full covariance matrix definitions would remain the same. A few of us had been quite confused over tracker behaviour until realising our model covariance inputs were variance and not standard deviation. Perhaps it could be documented somewhere?

[jmbarr]

Agree it would be worthwhile adding a line to the Kalman filter tutorial where the prior covariance ($P_0$) is first introduced.