Better error message when bin doesn't exist

[TApplencourt]

Reported by @zippylab

Before:

applenco@aurora-uan-0011:~> iprof adasd
Traceback (most recent call last):
	5: from /opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:840:in `<main>'
	4: from /opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:649:in `trace_and_on_node_processing'
	3: from /opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:309:in `open'
	2: from /opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:660:in `block in trace_and_on_node_processing'
	1: from /opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:424:in `launch_usr_bin'
/opt/aurora/24.180.0/spack/unified/0.8.0/install/linux-sles15-x86_64/oneapi-2024.07.30.002/thapi-git.ceaabfca6a67a0714c44fea8c49d24fc00559485_master-2il6zxsni7z7edscone7b2py6zeztzyr/bin/iprof:424:in `spawn': No such file or directory - fork failed (Errno::ENOENT)

Now

applenco@x4715c3s4b0n0:~/THAPI/build/ici/bin> ./iprof ze_infoads
./iprof: Can't find executable ze_infoads

Error message copied from mpirun

mpirun -n 1 adasdsd
Can't find executable adasdsd

Code a little ugly to my taste, but was not smart enough to do better. (and fixed a old bug of ensure)

[TApplencourt]

Pushed, the fix to call sync_d all the time.
This will avoid hanging is some weird people are playing with MPI_RANK_ID to launch different binaries, and one rank doesn't have the binaries (and that they don't use MPI, because if not mpi init will hang in there code, and nothing we can do about it).

[TApplencourt]

Will open another pr with the Sync_daemon renaming (and the yield f) after this one is merged to avoid rebase/conflic