shuf: use reservoir-sampling for large or unknown sized inputs

shuf: use reservoir-sampling for large or unknown sized inputs

Reservoir sampling optimizes selecting K random lines from large or
unknown-sized input: http://en.wikipedia.org/wiki/Reservoir_sampling
Note this also avoids reading any input when -n0 is specified.

* src/shuf.c (main): Use reservoir-sampling when the number of output
lines is known, and the input size is large or unknown.
(input_size): A new function to get the input size for regular files.
(read_input_reservoir_sampling): New function to read lines from input,
keeping only K lines in memory, replacing lines with decreasing prob.
(write_permuted_output_reservoir): New function to output reservoir.
* tests/misc/shuf-reservoir.sh: An expensive_ test using valgrind to
exercise the reservoir-sampling code.
* tests/local.mk: Reference new test.
* NEWS: Mention the improvement.