Work around slow path searches on MacOS

[fhunleth]

If a directory under /home was in $PATH on MacOS, the path search for
sed was surprisingly slow. One an M1 MBP, adding /home/missing to
$PATH caused mix firmware to regularly take >40 seconds. Without that
in the $PATH, mix firmware almost always takes 10 or 11 seconds.

See the conversation at
https://elixirforum.com/t/mix-firmware-takes-very-long-time-to-run-how-to-debug/41201
for much, much worse experience on a different MBP. A huge thanks to
jpalley for figuring out that the slowdown was correlated with the
/home directory in the path.

This PR caches the absolute path to sed to avoid searching $PATH each
time. On an M1 MBP, this reduces the mix firmware time from 40 seconds
to about 11 seconds.

[jpalley]

Nice job on the fast turnaround! (I'm the guy from the forum). FWIW:

• I was doing a bit of experimenting, and I think you can substantially speed up mix firmware, even when "/home" is not an issue by, by not forking sed 3+ times for every file. From my testing, these forks are a significant amount of the overall run time. I don't feel I know enough to pursue this (given it's day 1 for me with nerves!) but I was thinking an approach where unsquash_to_psuedofile and find_to_psuedofile basically sort the files objects into different lists for files/dirs/symlinks/etc. and then call a sed or awk program once for each list would see meaningful speedups.

• This is definitely a MacOS issue and nothing related to other stuff in this project. Stick something like this, "time echo "hi" | sed G" into a .sh file and run it with and without /home/ in the path. I see a 4x difference in "real" time to execute.

BTW - Loving nerves so far! Really slick.

[fhunleth]

@jpalley Thanks. I totally agree. The amount of calls to sed is completely over the top. There are so many things with this script that this is now really bugging me that there may be a significant speedup to mix firmware out there.