Add chunked encoding of large directories

[klauspost]

Above 25K entries do separate chunks to make FindSerialized able to search without full decompression, giving it a linear time characteristic. Minor compression loss (in artificial test set 1 byte/entry).

BEFORE:

BenchmarkFindSerialized/100-32         	   84632	     14027 ns/op	        24.28 b/file	         84632 files/s	    5062 B/op	       4 allocs/op
BenchmarkFindSerialized/1000-32        	   13371	     88087 ns/op	        23.32 b/file	         13371 files/s	   49398 B/op	       4 allocs/op
BenchmarkFindSerialized/10000-32       	    1491	    769408 ns/op	        20.66 b/file	          1491 files/s	  451106 B/op	       4 allocs/op
BenchmarkFindSerialized/100000-32      	     163	   7350855 ns/op	        20.25 b/file	         163.0 files/s	 4499216 B/op	       5 allocs/op
BenchmarkFindSerialized/1000000-32     	      15	 175437187 ns/op	        20.16 b/file	         7.500 files/s	44978624 B/op	       6 allocs/op

AFTER:
BenchmarkFindSerialized/100-32         	   88186	     13926 ns/op	        24.28 b/file	     88186 files/s	     216 B/op	       4 allocs/op
BenchmarkFindSerialized/1000-32        	   14157	     82875 ns/op	        23.32 b/file	     14157 files/s	     220 B/op	       4 allocs/op
BenchmarkFindSerialized/10000-32       	    1562	    757547 ns/op	        20.66 b/file	      1562 files/s	     225 B/op	       4 allocs/op
BenchmarkFindSerialized/100000-32      	     567	   2101872 ns/op	        21.66 b/file	       567.0 files/s	 1164419 B/op	      10 allocs/op
BenchmarkFindSerialized/1000000-32     	     583	   2035905 ns/op	        21.73 b/file	       583.0 files/s	 1181472 B/op	      47 allocs/op