feat: renumber cycles for discharge-first full cells (cellpy#359 / Stage 5 S3)

[jepegit]

Summary

Stage 5 S3 — mirrors jepegit/cellpy#359 (labelled to core): full cells that start with a lone discharge get a cycler cycle counter that mis-pairs half-cycles ([D C D][C D]… or [D C][D C]…), so coulombic efficiency and curve extraction are wrong.

Changes

• config.StepDirection (CHARGE / DISCHARGE) — the direction that opens a cycle. Kept separate from TestMode, which stays binary (polarity only).
• summarizers.renumber_cycles(data, schema=None, opening=StepDirection.CHARGE, **step_table_kwargs) — opt-in post-processing step, exported top-level:
  • per test_id, a new cycle opens where an opening step follows the last non-neutral step of the other direction; the first step is never a boundary, so a leading lone discharge stays cycle 1; rest / IR / OCV steps attach to the current cycle;
  • raw cycle-cumulative capacity / energy columns are re-accumulated to the new boundaries via _reaccumulate_cumulative (body extracted from normalize_capacity_granularity, which now calls the same helper);
  • exact no-op (early return) when the counter already matches; otherwise the step table is rebuilt with make_step_table(**step_table_kwargs) and a stale summary is dropped;
  • ValueError for ustep step tables / duplicate (test_id, cycle, step) keys, NoDataFound without raw or steps.
• Docs: standalone-use.md section, harmonized-raw.md cycle_num open question resolved, design note .issueflows/04-designs-and-guides/cycle-renumbering.md.
• Default engine behaviour unchanged; golden / e2e fixtures untouched.

Tests

tests/test_renumber_cycles.py (14 tests): the three patterns from the issue against a plain-Python oracle, rest/CV attachment, anode mirror (opening=DISCHARGE), 0-based numbering preserved, energy columns, two test_ids renumbered independently, pandas in/out, summary drop, error paths, and the harmonized anode fixture ([D C]x17 [D]): byte-identical no-op with opening=DISCHARGE, regrouped as [D] + [C D]x17 with capacity conserved under opening=CHARGE.

Full suite: 303 passed; ruff check / ruff format --check clean.

[D] [C D] [C D]        -> [D] [C D] [C D]        opening=charge   (no-op)
[D C D] [C D] [C D]    -> [D] [C D] [C D] [C D]  opening=charge
[D C] [D C] [D C]      -> [D] [C D] [C D] [C]    opening=charge
[C D C] [D C]          -> [C] [D C] [D C]        opening=discharge

Notes

• The cellpy/cellpy-core mirror issue could not be created from this environment (read-only token); the local tracking number 359 is provisional.
• cellpy-side wiring (a cycle_mode variant mapping to TestMode.NORMAL + renumber_cycles) and the release note are follow-ups in jepegit/cellpy.
• Changelog entry and version bump (minor) are left for /iflow-close.