Moves check for validity of resource.client from generic to specific instantiation of
Service::Prepare

Temporarily use submodule master until we are ready to commit to CTA master. Then we will restore
submodule updates from stable branch.

These files have been moved to a separate project in gitlab:

https://gitlab.cern.ch/eos/xrootd-ssi-protobuf-interface

Fixes failure mode where sizeof(uint32_t) + buflen exactly equals the
buffer size, and m_split_buflen had been set to a non-zero value in a
previous iteration.

Both tapeFile::ReadSession and tapeFile::WriteSession start by
reading VOL1 by calling Drive::readBlock() with a block of 84
bytes even though Logical Block Protection (LBP) is turned off
at this early stage of a session.  With LBP turned off, a drive
will only return the exact record being read and wil not add 4
CRC bytes at the end.  This means the drive will always return
80 bytes.

Both tapeFile::ReadSession and tapeFile::WriteSessioas called
Drive::readBlock() with a block of 84 bytes in order to work
with an "in-house" version of mhvtl that tried to support LBP.
Unfortunately the behaviour of this special version of mhvtl
was incorrect because it would 84 bytes when LBP was disabled.

The "in-house" version of mhvtl trying to support LBP is no
longer used with CASTOR.  This commit therefore simplfies the
CASTOR code by replacing Drive::readBlock() with
Drive::readExactBlock and by removing the conditional logic
that dealt with the existance of an 84 byte VOL1 label in the
presence of the "in-house" version of mhvtl.

Prepare was failing in client_ar.sh because of a typo in KrB5CCNAME, adding -s to prepare as by xrootd definition we are staging files to disk and not discovering them.

moment and won't be used in production (egroups).

Must add prepare acl on /eos/ctaeos/cta folder so that the retrieve works for the first test (may not be useful...)