Add measures for multi-targets

[ablaom]

I suppose there are two types of composition here - homogeneous and inhomogeneous, which might be implemented differently:

• homogeneous: all columns of target have the same scitype and we are we are defining our measure as a n-product of a single fixed uni-target measure m. Needed for time-series.

• inhomogeneous: any two measures (multi-target or uni-target), one can take their product m1 x m2, to get a new multi-target measure .

We should keep in mind that multi-target predictions are tables.

[ablaom]

Also, one may want to specify per-dimension weights, eg for weighting time series forecasts differently depending on how far in future the forecast is.

[ablaom]

Note that user-defined measures are a temporary workaround. See one suggestion posted here