[Austin243]

Adds reactionflow export-electrace <pathway> --out <folder>, which lays out one refined band for ElecTrace. Until now each band was laid out by a project script, as prepare_electrace.py did for the ACN density runs.

reactionflow export-electrace runs/<trajectory>/pathways/<occurrence> --out band-dft

band-dft gets:

• one folder per image (00, 01, ...), each holding only a POSCAR;
• electrace.toml, which reads the folders as one NEB path with atom_identity = "index". It sets pressure_gpa for an NPT band, so that ElecTrace compares H = E + PV at the band's pressure, the same enthalpy as barrier_eV (since Use the enthalpy for barriers at a stated pressure ElecTrace#23). focus_atoms lists the atoms of the bonds that differ between the band's two ends;
• reactionflow.json, with the pathway's result.json, the changed bonds as atom IDs and as POSCAR indices, and atom_ids, the stable ID of the atom at each POSCAR index.

Next, run the calculations and descriptor analyses in the frame folders (by hand or with ElecTrace's workflow), then electrace analyze band-dft/electrace.toml.

Design choices

• Atom order. ReactionFlow keeps bonds by stable atom ID. Every image is sorted by element in one stable order, so each species is one POSCAR block and an atom has the same index in every frame. Focus atoms are mapped through that order.
• No energies in VASP files. The model's energies and enthalpies stay in reactionflow.json, so ElecTrace never mixes MLIP energies with those of the calculations run on the frames.
• No new acceptance gate. A band that did not converge or failed its saddle check is still exported, and its status and saddle-check status are printed and recorded.
• Refusals. The export refuses:
  • a band whose images do not share their atoms and order;
  • an image with no cell;
  • an output folder that is not empty, so DFT results are never overwritten;
  • an event refined in steps, which has no band of its own. The message names its step-N folders, and each step exports with its own bonds.

Checks

Tested on a fresh run of examples/xtb (GFN2-xTB, 20 ps), which gave seven saddle_validated O–H–O proton-transfer bands:

• Every frame holds its image's atoms in the recorded order. Each element is one block (C H O), and no selective dynamics are written.
• The focus atoms are H3, O11 and O12. Those O–H distances go 1.00 → 1.71 Å and 1.72 → 1.00 Å across the band.
• ElecTrace loads the configuration, and electrace analyze runs on the exported folder. With no descriptors or energies yet, it warns about exactly that.
• No OUTCAR or OSZICAR is written, and the NVT band has no pressure_gpa.
• On copies of the band, edited for each case:
  • an NPT copy writes and loads pressure_gpa = 35.0;
  • a ci_neb_failed copy exports with its status reported;
  • a stepwise copy points to step-1, and step-1 exports with the same focus atoms;
  • images with a shifted atom order, a zero cell, or a non-empty --out are refused.
• All seven bands of the run export.
• ruff check, ruff format --check, pytest (356 passed, 25 skipped) and python -m build pass.