Add Eshelby memo, caching and lazy fields

[chemiskyy]

Introduce eshelby_memo to memoize Eshelby tensors for identical inclusion geometry/medium and reuse them within a homogenization pass (bitwise key). Integrate the memo into ellipsoid_multi: add S_loc_memo, eshelby_memoized(), and extend fillT/fillT_iso/fillT_mec_in to accept an optional memo; implement find/store/clear in C++ and use memcmp for exact-medium keys. Create a per-pass memo in various micromechanics schemes so identical integrations are shared, improving performance.

Python-side optimizations: make SolverResults compute expensive derived field 'LogStrain' lazily (pending derivations, _materialize, and persistent behavior on save/load/to_dataframe). In tensor.py add Basis pickling support that drops cached handles, add cached rotation handles and optional C++ rotation handles propagation to rotation paths to avoid re-creating C++ rotation objects in single-tensor rotations.

Other changes: rows_to_arr now moves the Armadillo matrix into the numpy conversion to avoid an extra copy; add required includes (vector, cstring, utility). Add tests (test_lazy_derivations.py and Teshelby test update) covering memo behavior, Basis cache/pickling, and lazy LogStrain materialization. Overall the changes reduce redundant recomputation and improve performance while preserving numerical results.