Fixing segfault crashes when using measure/reset ops.

[1tnguyen]

Description

This is probably a regression since #167

Specifically, flushRequestedAllocations() is not called when handling measurement at the concrete ExecutionManager implementation (i.e., QIRExecutionManager) with the assumption that it has been called during executeInstruction().
This will lead to a Segfault in this case (no gates)

{
        cudaq::qvector q(2);
        mz(q);
 }

For reset gate, there is a subtle issue w.r.t. gate ordering (leading to segfault eventually):
ExecutionManager::resetQudit is skipped by the middle BasicExecutionManager (where instruction batching occurs) and only implemented by the concrete subclass QIRExecutionManager -> reset op could be dispatched out of order due to no synchronize() call (-> segfault due to no flushRequestedAllocations call).

Changes

• Rename ExecutionManager::resetQudit -> ExecutionManager::reset and use BasicExecutionManager::resetQudit as the pure virtual method for sub-class implementation. This is to make it consistent with how measure is handled.

• Add flushRequestedAllocations() call to measureQudit/measureSpinOp/resetQudit implementations.

• Add unit test cases.

[amccaskey]

Thanks Thien!