B2: port concat_summaries feature set onto collect options (#706)

[jepegit]

Epic B, arc B2 (#706) — the collectors feature port

Carries the meaty three feature families of the legacy helpers.concat_summaries onto the SummaryOptions model built in B1, polars-native and using the cell's schema headers (the legacy headers_summary / headers_step_table singletons are deprecated-for-removal in 2.1).

Feature families

• Rate filtering — keep only cycles whose rate_on step ran at the requested C-rate (rate / rate_on / rate_std / rate_column / rate_inverse / rate_inverted). Ported from select_summary_based_on_rate as a polars step mask returning surviving cycles.
• Group averaging — average per journal group into a tidy long frame (group, cycle_num, variable, mean, std). Ported from _make_average + collect_frames; keeps the "<2 cells per group disables grouping" guard and names the aggregate mean regardless of average_method (backward-compat).
• CV partition — split each numeric metric into *_non_cv / *_cv via make_summary(exclude_step_types=["cv_"]). Ported from _partition_summary_based_on_cv_steps; column selection expands to keep the derived split.

Plus max_cycle / remove_last / only_selected cycle selection, normalized (equivalent) cycle exposure, and inf/extreme cleanup.

Layout

• collect/_summary_ops.py — per-cell polars ops (rate mask, CV partition, extraction, cleanup, group average).
• collect/summary.py — orchestration: per-cell extract → combine → select → group-average → cleanup → transforms.
• collect/options.py — SummaryOptions grown to the full pipeline, one source of truth (replaces ~30 concat_summaries kwargs).

Tests

tests/test_collect.py — golden tests for max_cycle, rate filter (+ inverted), CV columns, and group averaging (mean/std values checked). Full local run: 52 passed, 5 planned xfails (B3/B4 collectors placeholders).

Closes #706

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.