zstd: Speed up when WithDecoderLowmem(false)

[klauspost]

Typically an improvement when limited by the execute stage (enwik9)

BenchmarkDecoderEnwik9/multithreaded-writer-32         	       1	1184100400 ns/op	 844.52 MB/s	13992400 B/op	     101 allocs/op
BenchmarkDecoderEnwik9/multithreaded-writer-himem-32   	       1	1095716200 ns/op	 912.65 MB/s	19551184 B/op	      63 allocs/op
BenchmarkDecoderEnwik9/singlethreaded-writer-32        	       1	1612467600 ns/op	 620.17 MB/s	 6602968 B/op	      29 allocs/op
BenchmarkDecoderEnwik9/singlethreaded-writerto-32      	       1	1631256700 ns/op	 613.02 MB/s	 6590776 B/op	      26 allocs/op
BenchmarkDecoderEnwik9/singlethreaded-himem-32         	       1	1338224000 ns/op	 747.26 MB/s	 8835640 B/op	      23 allocs/

BenchmarkDecoderSilesia/multithreaded-writer-32         	       5	 201332380 ns/op	1052.72 MB/s	 5130272 B/op	      74 allocs/op
BenchmarkDecoderSilesia/multithreaded-writer-himem-32   	       5	 201792940 ns/op	1050.32 MB/s	 3105158 B/op	      42 allocs/op
BenchmarkDecoderSilesia/singlethreaded-writer-32        	       4	 268667300 ns/op	 788.88 MB/s	 1648902 B/op	      19 allocs/op
BenchmarkDecoderSilesia/singlethreaded-writerto-32      	       4	 265867950 ns/op	 797.19 MB/s	 1642810 B/op	      15 allocs/op
BenchmarkDecoderSilesia/singlethreaded-himem-32         	       4	 264649325 ns/op	 800.86 MB/s	 1161658 B/op	       7 allocs/o