Add public-domain classic reading fixtures

[isomorphisms]

Adds a small source manifest for exercising IB on ordinary long-form reading before adding more specialized paper behavior.

The fixtures stay source-level rather than inventing a novel-specific browser object. They cover:

• Njál's Saga in George Webbe Dasent's public-domain translation, for a long episodic narrative with genealogy, law, and non-novel structure;
• Sherlock Holmes short-story boundaries and a conventional Holmes novel;
• nested narration in Frankenstein;
• dated epistolary structure in Dracula;
• many uneven chapters in Moby-Dick;
• War and Peace as a very large multi-book stress text;
• Pride and Prejudice as a conventional chaptered baseline.

Each row points to a Project Gutenberg source page and UTF-8 plain-text representation that Gutenberg identifies as public domain in the United States.

Musashi is intentionally not included. Eiji Yoshikawa's novel was serialized in 1935–1939, and its familiar Charles S. Terry English translation dates from 1981. The U.S. copyright status is therefore not suitable for a public-domain fixture corpus in 2026.

Christ Stopped at Eboli and Leonardo Sciascia's books are also useful known-reading cases, but they are not public-domain source fixtures. They can later enter through user-owned/local representations without changing the browser model.

This is deliberately only a fixture stub: no novel-only semantic type, fragment policy, bibliography machinery, model hook, or reader behavior is introduced here.