X/mR lane order honours an ordered Categorical factor (#144, PR 4)

[cnicholas]

Part of #144 (PR 4 of 4). Stacked on #151: the base is feat/phased-xmr-phase-stats, so the diff shows only this change.

Summary

• What: the canonical row order in DataPreparation.build_keys sorts an ordered pandas Categorical factor by its category codes instead of its labels. Plain string and numeric factors sort exactly as before: the parallel key is built only when an ordered Categorical is present.
• Why: a stage column with categories before < after already ordered the Xbar/S subgroups that way (the aggregation groups on the factor column), but X/mR lanes, phases and result.strata came out after, before, because the row order compared the raw labels. The staged Xbar/S docs tell analysts to use labels that sort or an ordered Categorical; the second option now holds on every chart.
• Scope: one sort in build_keys. Residuals that run along the lane-major stream (R2 in ADS 2/3) follow the new order for ordered-Categorical inputs, which is the point: the stream should run in stage order.

Contract / invariants

• Defaults unchanged for every input without an ordered Categorical factor: the original sort_values(['cell_key', 'obs_id']) line runs untouched; validation/xbar_s_sweep.py identical (0 differing lines); golden masters untouched and passing
• Residuals unaffected for those inputs; for ordered-Categorical inputs they follow the corrected row order (intended)
• Row/index alignment preserved — obs_id remains the tie-break within a cell; sort_key is still 0..N-1
• Output schema compatible
• Pinned error strings unchanged

Behaviour changes

• Studies whose factor is an ordered Categorical: X/mR lane order, phase order, lane_boundaries, strata and the row order of the analysis frame follow the category order. CSV inputs are never categorical, so the Bishop gate and every reference run are unaffected.

Methodology

• No methodology change — ordering only

Tests

• tests/test_ordered_categorical_lane_order.py (5): ordered before < after gives before, after on phased X lanes, strata, phased Xbar phases, the lane divider and the analysis frame; plain strings still sort after, before; sortable labels need no Categorical; two factors with one ordered sort stage-major then machine; rows within a cell keep source order
• pytest tests/ — 2638 passed, 7 skipped
• ruff check .
• mypy processbehavior — 14 errors, all pre-existing, none in data_preparation.py
• python validation/e2e_bishop_report.py — 1547 passed, 0 failed, 4 pending; docs/reference/validation.md unchanged

Notes

• Docs: CHANGELOG Fixed entry. The "Staged Xbar/S" section added in Phased limits on Xbar and S charts (#144, PR 2) #150 already says an ordered Categorical is honoured.
• With this PR the four parts of Phased (staged) limits for Xbar and S charts #144 are open: Extract the per-group Xbar/S estimator for phased charts (#144, PR 1) #149 (refactor), Phased limits on Xbar and S charts (#144, PR 2) #150 (phased Xbar/S), Phase column and phase_statistics on phased X/mR (#144, PR 3) #151 (X/mR parity), this one. Merge in that order; each retargets to main as its base merges.