s2: Improve speed with bigger output margin

[klauspost]

Use bigger output margin and utilize it to reduce memmove branching.

Before -> after (best of 10)

Compressing... 10737418240 -> 4455442050 [41.49%]; 2.964s, 3454.9MB/s
Compressing... 10737418240 -> 4455442050 [41.49%]; 2.805s, 3650.2MB/s

Compressing... 10737418240 -> 4562648848 [42.49%]; 860ms, 11907.1MB/s
Compressing... 10737418240 -> 4562648848 [42.49%]; 856ms, 11960.0MB/s

Compressing... 10065157632 -> 5540789702 [55.05%]; 1.55s, 6193.7MB/s
Compressing... 10065157632 -> 5540789702 [55.05%]; 1.53s, 6273.7MB/s

Compressing... 10065157632 -> 5916578242 [58.78%]; 983ms, 9765.1MB/s
Compressing... 10065157632 -> 5916578242 [58.78%]; 972ms, 9878.3MB/s

Compressing... 10000000000 -> 4129259395 [41.29%]; 1.967s, 4848.0MB/s
Compressing... 10000000000 -> 4129259395 [41.29%]; 1.857s, 5134.7MB/s

Compressing... 10000000000 -> 4761467548 [47.61%]; 1.077s, 8853.7MB/s
Compressing... 10000000000 -> 4761467548 [47.61%]; 1.07s, 8910.8MB/s