yes: slower than GNU 9.11

[oech3]

$ taskset -c 0 yes |taskset -c 1 pv >/dev/null
^[[A71.3GiB 0:00:02 [35.7GiB/s]
$ taskset -c 0 yes |taskset -c 1 pv >/dev/null
^C.5GiB 0:00:02 [31.2GiB/s]
...
$ taskset -c 0 yes |taskset -c 1 pv >/dev/null
^C.8GiB 0:00:01 [29.8GiB/s]

$ LD_PRELOAD=/usr/lib/libmimalloc.so taskset -c 0 gnu9.11/yes|taskset -c 1 pv >/dev/null
^C.7GiB 0:00:03 [31.9GiB/s]
$ LD_PRELOAD=/usr/lib/libmimalloc.so taskset -c 0 gnu9.11/yes|taskset -c 1 pv >/dev/null
^C.1GiB 0:00:01 [32.1GiB/s]
$ LD_PRELOAD=/usr/lib/libmimalloc.so taskset -c 0 gnu9.11/yes|taskset -c 1 pv >/dev/null
^C.1GiB 0:00:01 [37.1GiB/s]

mimalloc does not help our tee() based yes since pipe cannot use huge page.
On the other hand, we cannot unsafe vmsplice to use huge page.

[oech3]

cc @ArniDagur