fix(pdf): preflight source objects before parsing

[McanKul]

What

• scan object-header candidates before lopdf parses them and bound object count/nesting
• reject recursive /Length chains beyond the classifier budget
• decode object streams with per-stream and aggregate caps, then nesting-check their members

Why

This keeps crafted source PDFs from driving unbounded recursion or object-stream decoding in the read-only text classifier.

Verification

• cargo test --manifest-path src-tauri/Cargo.toml --lib (302 passed)
• cargo test --manifest-path src-tauri/Cargo.toml --lib source_content (45 passed)
• rustfmt --check and git diff --check

Scope

Read-only classifier only: no command, UI, mutation, release, or main changes. This intentionally does not complete #33; raw xref preflight and bounded operator parsing remain separate follow-up PRs.

Refs #33