Diffusion Monte Carlo trace anomaly in batched QMCPACK code

**Describe the bug**
A Diffusion Monte Carlo calculation using the batched (real build CPU) version of QMCPACK 3.16.0 on ALCF's Polaris resulted in nearly periodic dips in the DMC energy trace which were on the order of 5 Ha. This issue is not present in an otherwise identical calculation using the legacy (real build CPU) version of QMCPACK 3.16.9 on NERSC's Perlmutter.

**To Reproduce**
1. If necessary, install QMCPACK 3.16.0 to Polaris from [here](https://qmcpack.org/releases)
2. Run QMCPACK 3.16.0 on ALCF's Polaris using the attached input files in the location *bug_inputs/dmc_3160_batched_polaris* and the orbitals specified in the NOTE below.
3. Plot the energy trace which results using the command: qmca -t -q e *.s001.scalar.dat

NOTE: The QP2QMCPACK.h5 file containing the orbitals is in a public Google Drive link [here](https://drive.google.com/file/d/1uPhhA2inhJkfCHqM_i2E3InmQqZlmYUA/view?usp=share_link) due to the large size of 98 MB.

**Description of the system**
An 8-atom primitive cell of ferromagnetic monolayer CrI3 with more than 20Å of vacuum in the z-direction, with 38 up spin electrons and 32 down-spin electrons. Used ppn boundary conditions and 4,4,1 periodic images. The orbitals are a single-determinant of natural orbitals (NO's) obtained from three iterative cycles of CIPSI -> NO's and represented in a GTO basis set.

**Expected behavior**
It is expected to see the same trace problem as in the attached photo *bug_inputs/trace_dmc_3160_batched_polaris.png*. Compare this to regular legacy Perlmutter behavior (*bug_inputs/trace_dmc_3169_legacy_perlmutter.png*).

**Inputs and photos of issue**
[bug_inputs.zip](https://github.com/QMCPACK/qmcpack/files/11874024/bug_inputs.zip)