[flang][cuda] Avoid hlfir.declare verifier error when creating temps (#…

…89984)

When creating temporaries for implicit transfer, the newly create
hlfir.declare operation was missing some information like the shape and
the verifier was throwing an error. Fix it by making sure we have an
ExtendedValue when calling addSymbol to register the temp.

```
error: loc("cuda-data-transfer.cuf":67:22): 'hlfir.declare' op of array entity
with a raw address base must have a shape operand that is a shape or shapeshift
```

Thanks @jeanPerier for the advice!

FYI @ImanHosseini

flang/lib/Lower/Bridge.cpp

@@ -3794,7 +3794,9 @@ class FirConverter : public Fortran::lower::AbstractConverter {
           auto needCleanup = fir::getIntIfConstant(cleanup);
           if (needCleanup && *needCleanup)
             temps.push_back(temp);
-          addSymbol(sym, temp, /*forced=*/true);
+          addSymbol(sym,
+                    hlfir::translateToExtendedValue(loc, builder, temp).first,
+                    /*forced=*/true);
           builder.create<fir::CUDADataTransferOp>(loc, addr, temp,
                                                   transferKindAttr);
           ++nbDeviceResidentObject;

flang/test/Lower/CUDA/cuda-data-transfer.cuf

@@ -98,7 +98,7 @@ end
 
 ! CHECK: %[[TEMP:.*]] = fir.allocmem !fir.array<10xi32> {bindc_name = ".tmp", uniq_name = ""}
 ! CHECK: %[[DECL_TEMP:.*]]:2 = hlfir.declare %[[TEMP]](%{{.*}}) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.heap<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>)
-! CHECK: %[[ADEV_TEMP:.*]]:2 = hlfir.declare %21#0 {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Eadev"} : (!fir.heap<!fir.array<10xi32>>) -> (!fir.heap<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>)
+! CHECK: %[[ADEV_TEMP:.*]]:2 = hlfir.declare %[[DECL_TEMP]]#1(%{{.*}}) {cuda_attr = #fir.cuda<device>, uniq_name = "_QFsub2Eadev"} : (!fir.heap<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.heap<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>)
 ! CHECK: fir.cuda_data_transfer %[[ADEV]]#1 to %[[DECL_TEMP]]#0 {transfer_kind = #fir.cuda_transfer<device_host>} : !fir.ref<!fir.array<10xi32>>, !fir.heap<!fir.array<10xi32>>
 ! CHECK: %[[ELEMENTAL:.*]] = hlfir.elemental %{{.*}} unordered : (!fir.shape<1>) -> !hlfir.expr<10xi32>
 ! CHECK: hlfir.assign %[[ELEMENTAL]] to %[[BHOST]]#0 : !hlfir.expr<10xi32>, !fir.ref<!fir.array<10xi32>>