Corrupted text extraction for text-based PDFs

## Summary

Some text-based PDFs are correctly detected as `TextBased`, but the extracted text is heavily corrupted. Characters become separated, words are broken apart, columns are merged incorrectly, and the resulting markdown is unsuitable for downstream processing.

---

## Example

### Expected

```text

John Doe

Software Engineer

Experience

Senior Software Engineer

ABC Technologies

Jan 2022 – Present

```

### Actual

```text

J O H N D O E

S o f t w a r e E n g i n e e r

E x p e r i e n c e

S e n i o r S o f t w a r e E n g i n e e r

```

The issue names no files, tests, or entry points. Start by reproducing the corruption with representative text-based PDFs and trace the text-extraction path; done means characters and words remain intact, columns are handled correctly, and the resulting markdown matches the expected content.

Written by the indexing model from the issue text.