docs(xlsx): measure horizontal bar-chart category-label baselines, document blockers (#1621)

[developer0hye]

Summary

• Confirms native Excel quantizes a horizontal bar chart's category-label vertical baseline to a whole sheet point: the issue's own 6 expense-chart baselines, divided by the sheet's 0.78 fit-to-height print scale, are exact integers. Extends this to a second chart (income, a different row pitch) and three font sizes (8/10/14pt) — 18 measured native baselines total, all whole sheet points.
• Falsifies the obvious fix: neither a flat floor nor a flat round of office2pdf's own current continuous baseline reproduces native across both charts with one shared rule. floor matches every expense-chart row but only 1 of 5 income-chart rows; round matches every income-chart row but only 3 of 6 expense-chart rows.
• Documents a more promising, still-incomplete decomposition (native = round(row's geometric centre) + a coarsely size-quantized term, mirroring line_category_baseline_pt's already-shipped ascent rounding) that fits the income chart exactly across all three sizes, but is off by one sheet point on 5 of 6 expense rows for a reason not yet isolated — most likely the plot rectangle's own vertical edges needing independent whole-point snapping before dividing into row bands. A category-count probe (not yet run — more invasive than the size probe here, since it needs to rewrite both the category and value caches together) would confirm or rule this out.
• No runtime code changes — assets/validation/issue-1621/ only.

Related issue

Related: #1621, #1756

Testing

• python3 assets/validation/issue-1621/verify_measurements.py exits 0, confirming the whole-sheet-point quantization and the floor/round split against both native datasets.
• python3 scripts/probe_harness.py assets/validation/issue-1621/category-size-probe.json --backend office reproduces the 8pt/14pt income-chart native rows recorded in measurements.json.
• Documentation-freshness review completed (two inaccurate claims found and fixed across two passes before this commit: a miscounted baseline total, and a false "farthest from the rounding boundary" superlative).
• No .rs/runtime config changes — tests skipped per project convention for non-runtime changes.

Visual impact

• No rendered PDF change
• Rendered PDF change or visual evidence added
• Reason: Adds only assets/validation/issue-1621/ measurement data, a probe spec, and a verification script; no chart rendering or codegen path is touched, so there is no rendered PDF change to audit.

Checklist

• Commits include a Signed-off-by line
• PR scope contains one root cause
• Remaining converter or harness deviations each reference an open issue

🤖 Generated with Claude Code