feat: direct text/URL ingestion — skip the wiki intermediary

[nickmeinhold]

Problem

Engram currently requires content to live in an Outline wiki before it can be extracted into the knowledge graph. This adds an intermediary step that most learners won't set up. ivvi takes raw audio; many other tools take raw text, URLs, or file uploads.

Proposal

Support direct ingestion from multiple unstructured sources without requiring an Outline wiki:

1. Paste text — paste lecture notes, article text, or any content directly into an ingest screen
2. URL fetch — provide a URL, Engram fetches and extracts (web articles, documentation pages)
3. File upload — PDF, EPUB, plain text files
4. Clipboard / share sheet — on mobile (Make mobile apps with haptic feedback on fracture #48), receive shared content from other apps via OS share sheet

All paths feed into the existing ExtractionService → knowledge graph → FSRS pipeline. The Outline wiki becomes one input source among many, not a prerequisite.

Relationship to existing work

• feat: semantic linking across sources — podcasts, books, and offline discovery #75 (cross-source ingestion) covers podcasts and books
• Live audio capture issue covers speech input
• ExtractionService already handles the Claude extraction — this is about diversifying what feeds into it

Why this matters

The wiki requirement is the biggest assumption baked into Engram's architecture. Most learners don't maintain a wiki. Meeting them where their content already lives (browser, clipboard, files) removes the highest barrier to adoption.