Combine multiple uploaded files into one synthesized deck

[aalemayhu]

Feature request from our heaviest AI-converter user (user 18996), via email 22 Jul 2026. Sibling of #3917 (same email, distinct ask).

Problem

They upload several sources that cover the same topic (a book chapter, the lecture transcript, the slide deck) hoping the AI combines them. Today PrepareDeck converts each HTML file independently — one generateDeckInfo call per file, one deck per file — so the sources never inform each other. The user gets N parallel decks with overlapping cards instead of one deck that uses the transcript to fill gaps in the slides.

Current honest guidance (told to the user): upload the single densest source.

Idea

A combined-conversion mode: when one upload contains multiple content files, offer "one deck from all files" that gives Claude the sources together (or sequentially with accumulated context) so cards synthesize across them and overlap collapses at generation time instead of surviving as duplicates.

Constraints to weigh in the spec

• Token cost: concatenating sources multiplies input tokens on the most expensive path (AI converter is ~88% of current Anthropic spend); chunking already splits large single files. Cross-file context may need the do-not-repeat-fronts mechanism (buildTopUpInstruction) rather than raw concatenation.
• Interacts with Cross-conversion duplicate card detection for overlapping source uploads #3917 — solving synthesis at conversion time removes most of the duplicate problem for the multi-file case.
• Needs trio + spec before build; filed to capture the signal.