fix: metrics average over compared pairs and throw on empty input

[skymanbp]

Split out of #214.

mse/mae/r2/accuracy/logLoss divided by length truth while zipWith truncated to the shorter vector, and scored zero predictions as perfect. They now average over compared pairs and throw on none. The internal meanSquaredError divided by the longer length and indexed out of bounds when predictions ran longer; it now returns Nothing on mismatched or empty input.

Before:

ghci> mse VU.empty (VU.fromList [5,5,5])
0.0
ghci> mse (VU.fromList [2,2]) (VU.fromList [0,0,0,0])
2.0

After:

ghci> mse VU.empty (VU.fromList [5,5,5])
*** Exception: [ERROR] mse cannot be called on empty data sets
ghci> mse (VU.fromList [2,2]) (VU.fromList [0,0,0,0])
4.0

sklearn raises ValueError on both empty input and length mismatch.