[kavanase]

Summary

get_stability returns get_e_above_hull for every input entry, including those the correction or mixing scheme dropped, whose energies it never placed on the hull's scale. The mixing scheme (default GGA_GGA+U_R2SCAN) drops entries without a structure or entry_id, and duplicates of a structure, such as a user's own calculation of a known MP material. For these, get_stability either:

• gave a value on the wrong scale; e.g. a GGA copy of mp-13 (bcc Fe), 0.05 eV/atom above MP's GGA entry, got 5.99 eV/atom; or
• raised for the whole call: ValueError if the entry fell below the hull, KeyError if it had no POTCAR info.

The scheme anchors the hull on one functional (r2SCAN if every GGA(+U) ground state has an r2SCAN match, else GGA(+U)) and leaves that functional's energies as its compat_1/compat_2 corrected them. So now:

• dropped entries of the anchoring functional are corrected the same way and placed on the hull (negative if below it);
• other dropped entries, and entries the correction rejects, get e_above_hull=None, with one MPRestWarning listing them.

Breaking: e_above_hull can now be None, but only where the old value was wrong or the call raised.

test_get_stability_dropped_entries has been updated to explicitly test this behaviour. It uses synthetic Li-O hulls anchored on each functional. User entries: GGA and r2SCAN twins of MP's Li, a plain oxide ComputedEntry, a GGA+U entry MP2020 rejects, an entry below the hull, and one without POTCAR info. Both cases raise ValueError: No valid decomposition found on main, fixed now.
For test_get_stability: Before, it compared nothing for GGA_GGA+U_R2SCAN (its entry was a bare ComputedEntry, which the scheme drops), and only one entry, of the last chemsys, for the other thermo types. Now, in every sampled chemsys plus Ni-S (a GGA(+U)-anchored mixed hull), perturbed copies of MP's compounds, 10 meV/atom higher, must sit 10 meV/atom above the copied material on MP's hull. A He-Ne test case checks the no-diagram path (only elemental entries).

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)