Add literal occurrence strands over stable fragments

[isomorphisms]

Implements the deliberately narrow first operation from #54 without adding novel-specific or model-derived machinery.

• searches the authoritative source text with deterministic exact-string matching;
• maps each occurrence start coordinate to an existing stable fragment source range;
• records the matching fragment and its position on a supplied materialized strand;
• returns bounded before/after fragment-id windows directly from that strand rather than repeatedly following next links;
• refuses a context lookup when a stored hit position no longer points at the same fragment;
• keeps duplicate textual occurrences distinct while also exposing unique matching fragment ids;
• includes exact Unicode matching and a match that crosses a fragment boundary, so chunking does not redefine source search semantics.

For #54 the supplied strand is the source's document-order strand. The Strand type itself remains source-agnostic, preserving #52's requirement that other strands may cross source-document boundaries.

The smoke fixture uses synthetic text and the known-reading Sherlock fixture identity; no copyrighted source text is added. The richer alias/entity/theme/classification/interpretation layers described in #54 remain out of scope.

This follows the fragment/link/strand architecture described in #52: stable fragment identities and source coordinates are durable semantics; the document-order strand is the cheap contiguous traversal representation.