Leopard GF8: Write directly to output

[klauspost]

Replace work slices with output slices, so we write directly to output

Saves a memcopy of final result.

Before/after:

BenchmarkEncodeLeopard50x20x1M-32                    100          11191776 ns/op        6558.42 MB/s        1816 B/op          2 allocs/op
BenchmarkEncodeLeopard50x20x1M-32                    122           9737593 ns/op        7537.83 MB/s        3352 B/op          3 allocs/op