Generate sample documents for parser and pipeline testing

Those are zip-wrapped XML containers (Office Open XML, OpenDocument) — generating them in pure browser JS without a 300 KB+ library is awkward. They will land in a follow-up phase once we vendor a small writer (likely docx-templates or pptxgenjs). For now the six text formats cover most pipeline and parser-QA needs.

The data uses Russian and English first/last names, city names, departments and date fields — realistic enough to test name-parsing and address-normalisation, fake enough that no PII risk exists. Emails are user1@example.test (RFC 6761 reserved TLD), so they will never accidentally route to a real inbox.

Same seed → same bytes lets you check parser output against a known-good fixture in a regression test. If you want randomness, swap out the file in a follow-up generate call — the seed only locks the content, not the file identity.

Yes — download a sample, then run it through the corrupted-file-generator tool to get truncated / bit-flipped / magic-byte-zeroed variants. The combination is great for testing both happy-path and sad-path parsers in one suite.