validation fails on 1d stages test with temporal blocking on intel64 arch

> make clean; make -j stencil=test_stages_1d arch=intel64 mpi=0
> bin/yask.sh -ranks 1 -stencil test_stages_1d -arch intel64 -validate -no-pre_auto_tune -no-auto_tune -no-warmup -num_trials 1 -trial_steps 2 -outer_threads 1 -inner_threads 1 -ep 0 -mp 0 -no-allow_addl_padding -init_seed 1.0 -l 64 -b 24 -bt 2 -no-bind_inner_threads -force_scalar
...
Comparing output var(s) in 'test_stages_1d' to 'test_stages_1d-reference'...
** mismatch at C(t=0, x=56): got 427.000000; expected 433.000000
TEST FAILED on rank 0: 1 mismatch(es).

Triage:
* threading and MPI have been ruled out.
* failure does not occur with -bt 1.
* failure does not occur with arch=avx512.