Serve the pre-built PD for property_data / conventional_unit_cell queries

[kavanase]

Summary

Follow-up to #1104 / #1121. get_entries_in_chemsys serves MP's pre-built (website) phase diagram for the default mixed GGA_GGA+U_R2SCAN thermo type, which guarantees every entry shares one energy scale. That was gated on not (property_data or conventional_unit_cell): requesting either flag instead fell through to re-applying MaterialsProjectDFTMixingScheme locally, which is slower (two downloads plus a local re-mix), drops entries the scheme cannot place, can anchor on a different hull than MP serves, and is nondeterministic across processes (materialsproject/pymatgen#3113) -- (I'll be suggesting a separate fix for that issue to pymatgen).

The result was that the same chemical system returned three different answers depending on which flags you passed. For Si-O:

call	before	after
get_entries_in_chemsys("Si-O")	422 entries	422
... property_data=["energy_above_hull"]	419 entries, different set + energy frame	422, identical
... conventional_unit_cell=True	405 entries, different set + energy frame	422, identical

Now the served phase diagram is used in all three cases and decorated afterwards:
property_data fields are fetched with a single materials.thermo.search and merged into entry.data, and conventional_unit_cell reshaping is applied to the returned entries.

Bug fix: conventional_unit_cell mis-scaled the r2SCAN mixing correction

Previously, get_entries reshaped entries by mutating the raw thermo doc dicts, scaling the energy, the composition, and any n_atoms-bearing energy adjustment. A ConstantEnergyAdjustment — how the GGA(+U)/r2SCAN mixing correction is stored — has no n_atoms, so it alone stayed at its primitive-cell (no conventional) value while everything around it scaled, silently shifting corrected energies.

On get_entries("H-O", additional_criteria={"thermo_types": ["GGA_GGA+U_R2SCAN"]}, conventional_unit_cell=True), 6 of the 11 reshaped entries came back wrong:

mp-24043-r2SCAN     -5.187195 -> -5.288040   delta -0.100845 eV/atom  (cell ratio 3)
mp-1212478-r2SCAN   -5.125967 -> -5.214010   delta -0.088043 eV/atom  (cell ratio 2)
mp-697111-r2SCAN    -5.192768 -> -5.268401   delta -0.075634 eV/atom  (cell ratio 2)
...

The magnitudes are exactly correction_per_atom x (ratio - 1) / ratio, i.e. entirely explained by the unscaled constant adjustment. Reshaping now goes through one helper, _get_conventional_cell_entry, which scales adjustments with pymatgen's own EnergyAdjustment.normalize — correct for constant, composition and temperature adjustments alike, which avoids these mismatches.

This affects get_entries, get_entry_by_material_id and get_entries_in_chemsys.

Changes

• get_entries_in_chemsys: drop the property_data / conventional_unit_cell gate; attach property_data to the served entries post-hoc with one thermo.search call.
• get_entries: build the ComputedStructureEntry first, then reshape the object via the shared helper.
• Apply the reshaping once, after every branch, rather than forwarding the flag into the queries — so the mixing scheme still matches structures on the original cells, and every path gets the same (correct) scaling.

Tests

• New test_get_entries_in_chemsys_decorated_served_pd: asserts the plain, property_data and conventional_unit_cell calls return the same served entry set on the same energy scale.
• test_get_entries now also asserts the corrected energy per atom. The existing assertion used the uncorrected energy, which was scaled correctly (while corrected was now).

Checklist

• Google format doc strings added. Check with ruff.
• Type annotations included. Check with mypy.
• Tests added for new features/fixes.
• If applicable, new classes/functions/modules have duecredit @due.dcite decorators to reference relevant papers by DOI (example)

[kavanase]

Test failure seems unrelated to this PR?