Because the page boundary is one of the few structural facts a PDF genuinely records. Nothing in the file marks a real heading, so inferring one from font size would be guesswork, whereas the page break is certain — it gives you a reliable anchor to navigate and edit by.

Because a scan contains an image of text rather than text, so there is nothing to extract. Test any PDF by trying to select a line in a viewer: if the cursor will not select it, no extraction tool will find it. Recognising characters in an image is a different job needing a different kind of tool.

Generally not as tables. A table in a PDF is text positioned to look like a grid, with no rows or columns recorded, so converting to Markdown usually leaves the values without their structure. Expect to rebuild anything where the arrangement carried the meaning.

No, the conversion runs in your browser and the PDF never reaches a server. Worth knowing given what people convert — reports, contracts and papers are the common cases, and none of them needs to leave your machine to have its text read.