Legacy bridge: preserve test_id through steps/summary output; legacy-schema support for merge_data

[jepegit]

Summary

The OldCellpyCellCore legacy bridge strips test_id from the outbound
step/summary frames, which degrades multi-test (campaign-merged) objects to
cycle-only summary grouping. cellpy #507 works around it; these requests make
the workaround unnecessary and unblock campaign merges that keep original
per-test cycle numbers.

Details

1. Outbound test_id stripping. _native_steps_to_legacy re-projects to a
   fixed legacy column order (cell_core.py ~781-815) and the summary path
   filters to _legacy_summary_column_order (~928-1065); test_id is in
   NATIVE_ONLY_STEP / NATIVE_ONLY_CYCLE (mapping.py ~191-196, 227-231). The
   inbound side is fine: a raw test_id column passes through (exception
   lists, mapping.py ~129-144) and make_step_table groups on
   (test_id, cycle_num, step_num) (proven by the cycler_small golden).
2. Bridge summary rebuilds native steps from the stripped legacy steps
   (make_core_summary, cell_core.py ~1010-1012), so
   use_tid (summarizers.py ~792) is always False through the bridge —
   cycle-only grouping, no per-test windowing, and cross-test collision when
   cycle numbers overlap. cellpy #507's workaround: re-stamp test_id onto the
   legacy steps frame post-bridge (depends on the inbound step rename staying
   rename-only with no column filtering — please treat that as a contract or
   provide the fix below).
3. _add_legacy_summary_cruft (cell_core.py ~960-978) does global pandas
   cumsums (cumulated CE, shifted_, cumulated_ric) even when the engine
   windowed per test — these columns are wrong for multi-test objects.
4. merge_data(schema=<legacy Schema>) raises AttributeError —
   HeadersNormal lacks the native attribute names (datapoint_num,
   cycle_num, test_id), and passing the native schema instead silently
   skips offsets on legacy-named frames (_max_column returns 0 for missing
   columns). Either document legacy schemas as unsupported for merge/update, or
   add attribute aliases so bridge consumers can call
   merge_data/update_data directly (cellpy #507 mirrors the semantics in
   pandas instead).

Requests

• Carry test_id through the legacy step output (or have
  make_core_summary reuse the native steps it already computed instead of
  rebuilding from the stripped legacy frame).
• Optionally expose test_id in the legacy summary output behind a flag.
• Window _add_legacy_summary_cruft per test_id when present.
• Resolve the merge_data legacy-schema story (docs or aliases).
• Together these unblock renumber_cycles=False (original per-test cycle
  numbers) for cellpy's campaign merge ahead of the native-schema path
  (cellpy #511).

Refs

cellpy jepegit/cellpy#507 (campaign merge + workaround), jepegit/cellpy#511
(native schema opt-in), core tests test_golden.py (cycler_small composite-key
golden), test_merge.py / test_e2e.py (native multi-test invariants).