Cache the result of forerunner to a tmp file

[liuchengxu]

Close #98

When the result set is too large for the built-in sync filter, and the furerunner job is completed, save the results to a tmp file instead of always running the heavy command again, which could be 2x faster.

$ cd ~ && fd --type f > home.txt

# Reading from a file
$ time /Users/xlc/.vim/plugged/vim-clap/target/release/maple "srli" --input /Users/xlc/home.txt --number 96
real	0m1.674s
user	0m1.640s
sys	0m0.093s

# Runing the command and pip into maple
$ time fd --type f | /Users/xlc/.vim/plugged/vim-clap/target/release/maple "srli" --number 96
real	0m3.173s
user	0m6.310s
sys	0m13.075s

This commit 3557163 using rayon's par_lines gains 3x faster again than reading file using std lib.

$ time /Users/xlc/.vim/plugged/vim-clap/target/release/maple "srli" --input /Users/xlc/home.txt --number 96
real	0m0.491s
user	0m2.838s
sys	0m0.158s

In conclusion, reading from a cached file can be 6x faster then using the pipe command.