fix: replace move_alloc on patch_ib with allocate/copy for GPU safety

patch_ib has GPU_DECLARE(create=...) in m_global_parameters, which means
OpenACC/OpenMP tracks it via plain allocate/deallocate automatically.
move_alloc is not reliably intercepted by all GPU runtimes for declare-
create variables, so replace all three grow-on-demand sites (m_particle_bed,
m_mpi_proxy, s_reduce_ib_patch_array 1-rank/no-MPI) with the established
pattern: allocate tmp, copy, deallocate patch_ib, allocate patch_ib.

Author: sbryngelson

src/simulation/m_mpi_proxy.fpp

@@ -199,7 +199,9 @@ contains
                 type(ib_patch_parameters), allocatable :: tmp(:)
                 integer                                :: n
                 n = size(patch_ib)
-                call move_alloc(patch_ib, tmp)
+                allocate (tmp(n))
+                tmp(1:n) = patch_ib(1:n)
+                deallocate (patch_ib)
                 allocate (patch_ib(num_ib_patches_max))
                 patch_ib(1:n) = tmp
             end block

src/simulation/m_particle_bed.fpp

@@ -46,7 +46,9 @@ contains
                 type(ib_patch_parameters), allocatable :: tmp(:)
                 integer                                :: n
                 n = size(patch_ib)
-                call move_alloc(patch_ib, tmp)
+                allocate (tmp(n))
+                tmp(1:n) = patch_ib(1:n)
+                deallocate (patch_ib)
                 allocate (patch_ib(num_ib_patches_max))
                 patch_ib(1:n) = tmp
             end block

src/simulation/m_start_up.fpp

@@ -1241,8 +1241,10 @@ contains
 
 #ifdef MFC_MPI
         if (num_procs == 1) then
-            ! single-rank: every patch is local - transfer ownership directly to avoid truncation
-            call move_alloc(patch_ib_gbl, patch_ib)
+            ! single-rank: every patch is local; allocate to exact size and copy
+            allocate (patch_ib(num_gbl_ibs))
+            patch_ib(1:num_gbl_ibs) = patch_ib_gbl(1:num_gbl_ibs)
+            deallocate (patch_ib_gbl)
         else
             ! multi-rank: carve out the local neighbourhood subset
             num_aware_ibs = min(num_local_ibs_max*(2*ib_neighborhood_radius + 1)**num_dims, num_ib_patches_max)
@@ -1271,8 +1273,10 @@ contains
             deallocate (patch_ib_gbl)
         end if
 #else
-        ! no-MPI: every patch is local - transfer ownership directly to avoid truncation
-        call move_alloc(patch_ib_gbl, patch_ib)
+        ! no-MPI: every patch is local; allocate to exact size and copy
+        allocate (patch_ib(num_gbl_ibs))
+        patch_ib(1:num_gbl_ibs) = patch_ib_gbl(1:num_gbl_ibs)
+        deallocate (patch_ib_gbl)
 #endif
 
         @:ALLOCATE(ib_gbl_idx_lookup(1:num_gbl_ibs))