Fixed case where a gridFTP client sends overlapping chunks. We now raise an error instead of screwing the checksum

Fixed bug #104856: RFE: make the timeout to expire old requests different for failed vs. successful ones

The previous commit was a bit fast, this is now the correct refactoring without goto. Moreover, now we expose timeout errors as SETIMEDOUT, instead of overwriting serrno with a generic SEINTERNAL.

Removed infinite loop around stage_open(), introduced long ago (2005) to compensate for network errors. We now prefer to expose network errors and timeouts and give users the option to retry. At the same time, minor refactoring of that code that dropped a label and extended the error handling to the rfio_open() calls.

Fixed an 'uncaught exception' error that can happen when a canceled job had been handled by the backfill queue

Fixed bad checkForReconnection in teh transfermanagerd. We were checking far too often due to misindentation and python taking identation into account

Amended the release notes for 2.1.14-13 after verification that a transparent upgrade cannot be performed

constraint is enough, while for master we keep the new simplified code.

constraint violation.
Removed constraint and reworked filesDeletedProc. It is OK to
drop a DiskCopy entry with an outstanding Disk2DiskCopyJob, when
it will start the case will be handled properly.

constraint violation.
The fix is to differentiate the handling of the constraint violation
exceptions: on diskCopy, everything has to go, while on castorFile
no action is taken as the entry could have been reused. Thus the code
specifically ignoring the castorFile exceptions has been moved around
the DELETE FROM CastorFile clause.

        https://savannah.cern.ch/bugs/index.php?104488

        svcClass for TPC transfers.