Not all cases of ILLEGAL w.r.t. the dense descriptor are caught #388
Description
In between #379 and #382, a transient meta-bug was noticed where sometimes the scaling perftest failed. The culprit turned out to be the use of the dense descriptor (to an mxv) while the output vector was sparse (empty, in fact). It was transient because only the nonblocking backend would pick up on this error, but only as part of a type of race condition (see #379 or 1414c49 for details). It would have been better if all other backends (reference, reference_omp, hyperdags, bsp1d) were also picking up this error - especially since these are blocking mode backends they should have been able to catch this error deterministically.
This issue should review all user-facing primitives in all backends to ensure they return ILLEGAL when called with any sparse vector while the dense descriptor is also given.