test(regression): harden chart-cell and waterfall overlay coverage

[tmustier]

Summary

Stabilization-oriented regression hardening for high-risk chart behaviors:

1. Waterfall connector overlays (style/dash variants)
2. Chart template copy application path (immediate replacement)
3. Chart-cells grouped headers with sub styling
4. Horizontal chart-cell manual data-label offsets with negative/long values

Changes

tests/test_chart_engine_smoke.py

• Added _pptx_part(...) helper for direct PPTX part assertions.
• Added _blank_auto_shape_count(...) helper to count connector-like blank auto-shapes.
• Added _waterfall_overlay_spec(...) helper for connector style matrix runs.
• Added test_build_chart_applies_chart_template_copy_immediately_when_saving(...)
  • verifies chart_template_copy: true is applied when save=True and defer_template_copy=False
  • asserts output chart1.xml and chart1.xml.rels match template parts exactly
• Added test_build_chart_waterfall_connector_styles_change_overlay_segment_counts(...)
  • compares gap vs step
  • compares solid vs long_dash vs dot
  • asserts style changes produce expected relative segment counts

tests/test_chart_cells_golden.py

• Added slide XML helper (_slide_xml) + shape token finder (_find_text_shape_with_tokens).
• Added test_horizontal_chart_cells_keep_distinct_offsets_for_negative_and_long_values(...)
  • validates manual horizontal label offsets remain positive and non-collapsed under mixed sign/long values
• Added test_grouped_chart_cells_render_sub_headers_on_new_line_with_body_color(...)
  • grouped chart-cell scenario with row-group headers containing sub
  • asserts:
    • grouped labels/connectors still render
    • sub appears on a new line (a:br)
    • main run remains bold
    • sub run is non-bold (b != 1)
    • sub run uses default body color (tx1)

Validation

• .venv/bin/ruff check clean_slides tests
• .venv/bin/pyright
• .venv/bin/pytest -q
• .venv/bin/pre-commit run --all-files

All pass locally.