Nofib: Ignore cr-lineendings in benchmarks

At least the real/eff benchmarks fail because of this on Windows.

==nofib== CS: time to run CS follows...
../../../runstdtest/runstdtest ./CS  -o1 CS.stdout -o1 CS.stdout  -ghc-timing  -stdout-binary    ;   ../../../runstdtest/runstdtest ./CS  -o1 CS.stdout -o1 CS.stdout  -ghc-timing  -stdout-binary    ;   ../../../runstdtest/runstdtest ./CS  -o1 CS.stdout -o1 CS.stdout  -ghc-timing  -stdout-binary    ;   ../../../runstdtest/runstdtest ./CS  -o1 CS.stdout -o1 CS.stdout  -ghc-timing  -stdout-binary    ;   ../../../runstdtest/runstdtest ./CS  -o1 CS.stdout -o1 CS.stdout  -ghc-timing  -stdout-binary    ;

real	0m0.173s
user	0m0.000s
sys	0m0.015s
././CS < /dev/null
expected stdout not matched by reality
<<ghc: 640047168 bytes, 611 GCs (609 + 2), 0/0 avg/max bytes residency (0 samples), 290232 bytes GC work, 2M in use, 0.000 INIT (0.000 elapsed), 0.125 MUT (0.124 elapsed), 0.000 GC (0.001 elapsed), 0.000 GC(0) (0.001 elapsed), 0.000 GC(1) (0.001 elapsed), 1 balance :ghc>>