fix: consistent sync/exit and error management in pipeline contexts

fix: consistent sync/exit and error management in pipeline contexts

Don't clobber an exception on exit of the nested block too. In order to
simplify the code, make the pipeline count the number of time it is
entered, and call _exit() only the last time it exits.

Drop assert that we have left pipeline mode leaving the block. If we get
in unrecoverable state, we will have not. By now we should probably just
close the connection; however, leaving it this way is a better
indication that the connection is broken because of something about
pipeline mode; closing it would hide it, and even if we raised a
warning, it would be much easier to miss it than to miss the exceptions
raised in broken state.