Fix OUTCAR parsing for VASP 5.4.4 with selective dynamics: variable m…

[eliakim-k]

Fix OUTCAR parsing for VASP 5.4.4 (selective dynamics) and harden dielectric-tensor reading

Problem.
get_modes_from_OUTCAR assumes exactly 3xnat modes. With selective dynamics, VASP prints modes only for unconstrained atoms (3xN_free), so the fixed-length loop overruns the eigenvector block and crashes with AttributeError: 'NoneType' object has no attribute 'group'. Reproduced on VASP 5.4.4 (build Nov 2024) for a Cu(111) slab: nat=55, 48 frozen Cu atoms → only 21 modes exist, not 165.

Changes.

• get_modes_from_OUTCAR: parse however many f = frequency lines exist and auto-detect per-mode atom count from the file (cross-checked against POSCAR nat, warns on mismatch).
• get_epsilon_from_OUTCAR: extract the dielectric tensor via float-regex instead of str.split(), tolerating merged columns / signs / scientific notation.
• Added get_epsilon_from_vasprun_xml + get_epsilon fallback wrapper.

Note. Imaginary (f/i=) modes are skipped by the frequency detector (intended for off-resonant Raman of stable structures); mode indices then refer to real modes only.
Verified against a real OUTCAR: 11 real modes parsed with correct frequencies and the dielectric tensor read correctly.

[afonari]

Can you generate diff that only changes the actual code. Also how is this different compared to e309fa3