fix: avoid O(n^2) value lookups in PPTX chart conversion

[kz-000]

Problem

Converting a PPTX that contains a large chart is extremely slow and can appear to hang. A real-world deck with a 1097-point, 3-series chart spent ~2.7s just reading the chart values, and larger charts effectively froze the conversion.

Root cause

PptxConverter._convert_chart_to_markdown accessed series.values[idx] inside the nested (category × series) loop. In python-pptx, series.values rescans the cached data points via XPath on every access (O(n) per lookup), so the code was O(n²) per series and rebuilt the entire values tuple once per category — O(n³) overall.

Fix

Materialize each series' values once before the loop, reducing the lookups to O(n). The generated Markdown is unchanged.

Benchmark (3 series, varying category count; identical output verified)

categories	before	after	speedup
50	0.48s	0.01s	47×
100	3.12s	0.03s	99×
200	22.4s	0.11s	205×
400	164s	0.40s	407×

Tests

Added test_pptx_chart_multi_series_conversion, which builds a multi-series chart with many categories and verifies the resulting Markdown table (series headers, first/last categories, and a representative data row).

[kz-000]

@microsoft-github-policy-service agree

[afourney]

This looks quite sensible! Thanks for the fix.