Stabilize RTP electric-field ramp energy

[DCM-Uni-Paderborn]

Summary

Two Ehrenfest/RTP energy matchers sat just below their observed cross-build numerical variation after #5755 corrected diagonal second derivatives of overlap integrals.

• Update the H2-emd-efield-ramp.inp reference to the corrected serial-debug value, -0.870455433182, and use a 3e-11 relative tolerance.
• Raise only the tolerance of H2O-ls-emd-mixing.inp from 2e-11 to 3e-11; its reference remains unchanged. The optimized GNU ssmp CI result differed by 2.10e-11, just beyond the former limit.

These are matcher-only changes. No source code or regtest input is changed.

Validation

• Reproduced -0.870455433182 for the field-ramp test with a current psmp build.
• Built current master as GNU ssmp on Spark and ran H2O-ls-emd-mixing.inp five times each with 1, 2, and 4 OpenMP threads. All 15 runs remained within 1.3e-11 of the retained reference; the last digits vary slightly with threaded reductions.
• ./make_pretty.sh
• git diff --check

Current CI status

The two RTP tests changed here pass in the targeted sdbg and ssmp runs. The remaining failures are outside this patch:

• the platform-sensitive GFN2 modified-Broyden test added on master, addressed by Stabilize GFN2 modified Broyden regtest #5810
• one borderline TDDFPT_Check_Osc_Strength result in ssmp (1.01e-5 versus a 1.0e-5 tolerance), which passed unchanged in a subsequent ssmp run

No reported failure is attributable to the RTP matcher changes.

[DCM-Uni-Paderborn]

/cp2kci sdbg intel-ifx-ssmp intel-ifx-psmp intel-ifort-ssmp intel-ifort-psmp openmpi-psmp psmp-4ranks rawhide-pdbg

[DCM-Uni-Paderborn]

/cp2kci intel-ifx-ssmp intel-ifx-psmp intel-ifort-ssmp intel-ifort-psmp openmpi-psmp psmp-4ranks rawhide-pdbg

[DCM-Uni-Paderborn]

/cp2kci sdbg ssmp

[DCM-Uni-Paderborn]

/cp2kci sdbg ssmp