Execute body of host_task if main interpreter is still initialized

This PR applies the solution worked out in a toy POC

Single taskes are executed in a separate threads from
the main thread which submitted the kernel and incremented
reference counts.

When kernel submission occurs near the end of the script,
the CPython interpreter may have begun finilization process
by the time the kernel completes the execution.

PyThread_Ensure would crash in that case. So execute the host
task only if `Py_IsInitialized()`.

The only testing of this scenario is to submit the kernel at
the end and not call queue synchronization.

Author: oleksandr-pavlyk

dpctl/_host_task_util.hpp

@@ -55,7 +55,8 @@ int async_dec_ref(DPCTLSyclQueueRef QRef,
                     *(reinterpret_cast<sycl::event *>(ERefs[ev_id])));
             }
             cgh.host_task([obj_array_size, obj_vec]() {
-                {
+                // if the main thread has not finilized the interpreter yet
+                if (Py_IsInitialized()) {
                     PyGILState_STATE gstate;
                     gstate = PyGILState_Ensure();
                     for (size_t i = 0; i < obj_array_size; ++i) {