RFC: Refactor document upload pipeline

[wyuc]

The upload pipeline is hardcoded around PDF. Every variable, function, and storage key uses pdf* naming, and the parsing logic is 200+ lines inlined in generation-preview/page.tsx.

We have 6+ open PRs adding new formats or modifying this pipeline (#41, #44, #65, #140, #160, #191), and they all conflict with each other on the same file.

Plan

1. Extract the inline parsing into a parseDocument() function with a per-format parser registry (parse-pdf.ts, parse-text.ts, etc.)
2. Rename pdf* to document* across the pipeline (TypeScript only, no DB migration)
3. After this lands, adding a new format = one parser file + register the extension. No need to touch the generation pipeline.

What this means for open PRs

PRs in this area will need a rebase after the refactoring. The rebase should simplify your diff, not complicate it. Happy to help with that.

Roadmap after merge

Priority	Item
P0	MD/TXT support (#41)
P0	Context-aware web search (#258)
P1	DOCX support (#140)
P1	MinerU cloud API (#180/#191)
P2	Multi-file upload (#148/#160)
P3	Long document summarization (#52/#65)

Feedback welcome.