E3: remove headers_* attribute access + make_new_cell

[jepegit]

Epic E — Shim removals (BREAKING) · Header access + misc. DEPRECATIONS.md removal: 2.1.

Remove:

• legacy headers_normal/headers_summary/headers_step_table attribute access → c.schema.raw/steps/summary
• make_new_cell → CellpyCell.vacant

Startable immediately. Part of #696.

[jepegit]

E3 progress — scope corrected, runtime nearly done

Key correction: the deprecated surface is only the CellpyCell.headers_* property shim (LegacyHeaderShim, which warns + maps to native), not the module-level headers_normal/get_headers_* singletons the instrument loaders use for pre-boundary column naming. So E3 is a small, targeted migration (~a few dozen field-mapped accesses), not the ~167 a flat grep suggests.

• make_new_cell → removed (E3 (partial): remove make_new_cell (#715) #743, merged).
• Runtime part 1 (E3 (runtime part 1): migrate direct headers_* accesses to native (#715) #744, merged): the 10 direct property accesses — cellreader.py (ustep, datetime_txt), tabular.py + slicing.py (energy columns).
• Runtime part 2 (E3 (runtime part 2): migrate headers_* resolvers to native (#715) #745, armed): the assign-to-variable + family-resolver patterns — helpers.select_summary_based_on_rate, capacity_curves.get_ocv, plotting/registry.py summary family resolvers, prepare/summary.py, plotutils.py, and removed the context.headers_summary passthrough. 89 tests green.

Remaining before the shim properties can be deleted:

1. exporters/bdf.py — deferred: its _COLUMN_MAP is keyed by legacy _txt field names via getattr(headers, spec.cellpy_field), which the shim translates to native columns. Needs a dedicated map→native migration (~12 fields). (This was the one file where a naive swap broke — the exporter resolves fields generically.)
2. Property removal — delete the LegacyHeaderShim wiring in cellreader.py (self.headers_normal/summary/step_table = _shims[...]), trim the DEPRECATIONS row, and migrate the ~110 test sites (most of which are also non-deprecated singleton usage and will collapse).

[jepegit]

✅ Runtime migration complete (parts 1–3 merged)

• make_new_cell removed (E3 (partial): remove make_new_cell (#715) #743).
• Runtime part 1 (E3 (runtime part 1): migrate direct headers_* accesses to native (#715) #744): direct property accesses (cellreader ustep/datetime_txt, tabular/slicing energy cols).
• Runtime part 2 (E3 (runtime part 2): migrate headers_* resolvers to native (#715) #745): assign-to-variable + family-resolver patterns (helpers, capacity_curves, plotting registry/prepare/plotutils, context passthrough removed).
• Runtime part 3 (E3 (runtime part 3): migrate bdf exporter off the headers shim (#715) #746): bdf exporter _COLUMN_MAP rebased onto native raw column names.

No non-loader runtime code accesses the CellpyCell.headers_* shim anymore. The test_core_pipeline_does_not_trip_its_own_deprecation guard confirms the processing pipeline is shim-free.

Final step (one PR): remove the shim

1. Migrate the ~30 remaining test-site accesses (cell.headers_* on a CellpyCell) across 12 test files to native schema.* / resolved literals — most of the raw ~110 count is non-deprecated get_headers_*() module-singleton usage that stays.
2. Rewrite test_cell_schema.py's shim characterization tests (test_legacy_shim_and_schema_agree, test_legacy_attribute_warns_and_names_the_schema_replacement) as "removed" checks.
3. Delete the LegacyHeaderShim wiring in cellreader.py (self.headers_normal/summary/step_table = _shims[...]), trim the DEPRECATIONS row, regenerate DEPRECATIONS.md.

[jepegit]

Property removal attempted — deferred as a focused final step

Went to remove the shim (full removal, both paths per decision): removed the cellreader.py wiring, deleted legacy_header_shim.py + its test, and migrated the straightforward direct-attr test sites. But the remaining test migration has genuine per-test subtleties that shouldn't be rushed:

• step.info and step.ustep are legacy-only columns with no native counterpart, used in test_cell_readers.py's step-table characterization assertions — those need per-test rethinking, not a mechanical rename.
• test_ocv_relax.py iterates dataset.headers_normal.items() (a diagnostic loop over the header object) — needs rewriting, not swapping.
• ~100 downstream attr renames across ~8 test files, several in loader/step characterization tests where intent matters.

Rather than land a rushed, half-verified migration on a red tree, I reverted the removal — the shim stays intact (green) and the runtime migration stands as the complete E3 deliverable.

The property-removal final step is now precisely defined: delete the cellreader.py shim block (both branches) + legacy_header_shim.py(+test), migrate the ~30 test-site accesses (native map recorded above), handle the step.info/step.ustep legacy-only characterization tests + the .items() loop deliberately, rewrite test_cell_schema.py's shim tests, and trim DEPRECATIONS. A clean, bounded focused task.