Fix parsing of centered borderless tables

[ShiyuanHsu]

Summary

• Preserve borderless table regions through XY-cut when rows share stable centered column tracks.
• Let the existing Markdown table classifier process the intact rows.
• Add regression tests for two-row tables and protection against vertical table slicing.

Root cause

For centered cell text, varying text widths change the left edges while the column centers remain aligned. XY-cut interpreted the whitespace between those columns as page-column gutters and split one logical table into separate regions. The Markdown classifier then received column-wise text instead of complete table rows.

Validation

• cargo test -p liteparse --no-default-features --lib centered_text_table_obstacle -- --nocapture
• Manually verified local parsing of native-text PDF samples without OCR.

[logan-markewich]

For changes to markdown, I would recommend re-running ParseBench -- we have tracked results there so its worth checking if this impacts performance there. I can always run it too if thats too annoying. I usually just point claude code or codex at it and let it wire up a local build for testing