Some of the glibc intercepted functions,
that are accepting the directory file descriptor,
where still using legacy code to relativize the input path.

If we intercept function during shutdown we could try to access the
internal file map that has been already freed.

This woun't affect performance at runtime since we are always checking
only one boolean flags on every interception as before.

On the other hand should sppedup the shutdown of the library, because we
disable the interception of glibc library call before performing the
shutdown routines.

Since mercury fixed the issue on their code [1], we can drop the
temporary patch we were using to avoid handles reset.

Due to the mercury fix The follwing two issues have been solved:
 - Mercury: https://github.com/mercury-hpc/mercury/issues/272
 - Margo: https://xgitlab.cels.anl.gov/sds/margo/issues/47

[1]: https://github.com/mercury-hpc/mercury/commit/9512b8b04c265645f95fd99f3b9c015e836c0347

 - improved statfs system call
 - intercepted statvfs library call

Both of the above function now report the aggregated real size
of all the data-nodes.

You can test this with `df "/tmp/gekkofs_mountpoint"` command

mercury changed again its interface.
Before was `ofi+sockets://fi_sockaddr_in://<IP>:<port>` but now
it only works if we omit the intermediate protocol `ofi+sockets://<IP>:<port>`

They fixed a potential race condition:

https://github.com/mercury-hpc/mercury/commit/a421ea3072c79b2edb6d4d9d2bdaf179b5ef14f3

See: https://xgitlab.cels.anl.gov/sds/margo/issues/47

Workaround must be reverted once fixed.

Changes have been merged upstream [1] with commit `acb8c296a694d7e3e06756e7a102a9d1c92df0c8`.
There is no more need to patch margo to expose enanched init function.

[1]: https://xgitlab.cels.anl.gov/sds/margo/merge_requests/9

since back() has undefined behaviour on empty string we need to ensure
that the string is not empty before to check the last char.

Paths passed as parameter to the daemon needs to be canonicalized

`gflags` is an optional dependency of RocksDB, it is used only for cli
tools that we don't use.

Since now there is only one margo instance there is no more difference
between RPC and IPC.

All the "ipc" occurence in the code have been removed in favor of "rpc"

removed `MARGO_FORWARD_TIMER` option that allows to measure RPCs timing