Sort 'rpc' exception handling can cause deadlock

In testing hpcc-8905, I noticed that exceptions caused during some of the sort master->slave communication calls, could cause those exceptions to never be reported.

In particular, there are a number of async calls that if they fire an exception, will be trapped, ready to be reported, but never reported back to anything.

The SORT master<>slave communication really needs refactoring, the structures in place now are a hang over from when the calls were hrpc calls over socket and they are convolution the process.