Add read-ahead buffer to decompress in parallel.

Add read-ahead buffer to decompress in parallel.

Allows using more than one CPU core for a good speedup.
Benchmarks:
Uncompressed files:    196.29 CPU, 5:18.34 wall clock time
xz-compressed, before: 299.19 CPU, 5:21.85 wall clock time
xz-compressed, after:  308.96 CPU, 3:29.60 wall clock time

(first was I/O limited, second was CPU-limited, now it is
almost only limited by CPU-time for XML parsing)