fix(sepia): compose author and executor prose layers; tag Mythos 5.1 as in scope

Two Codex findings on #45, round 2.

Composition. The identity rule selected both families' prose layers
when author and executor differ (author=Claude, executor=GPT) without
saying how they combine, and the report formats allowed one
"Prose layer:" value, so one family could go unreported. The rule now
states the two layers do not compete — the author's layer is hunted in
the text you were given, the executor's layer governs the text you
produce — and the professional-pass and rubric report lines record one
status per loaded family ("Claude=operative GPT=prior").

Scope tag. research/sources.md records the Anthropic page as covering
Claude Fable 5.1 and Claude Mythos 5.1, but the Claude prose-layer table
was tagged Fable 5.1 only, so an exact-match rule would have downgraded
a Mythos 5.1 executor to a prior. The table heading and the row-1
trigger now name both releases.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TtoUEQarAYtQYrZLLwqjao