seed.yaml awareness for manuscript metadata

[Shinran7]

ProblemGhostreader analyzes manuscripts without knowing the author's intent. It can't distinguish a 5-chapter short story from an incomplete novel, or know that high dialogue density is intentional.## SolutionRead the seed.yaml file (Autonomicon format) from the manuscript directory or its parent. Use the meta block to inform agent prompts and calibrate analysis.### Useful fields from seed.yaml meta- genre -- auto-set genre lens (currently requires --genre flag)- target_chapters / target_chapter_length -- avoid false "underdeveloped" flags- arc_archetype -- evaluate structure against intended arc (e.g. long_defeat shouldn't be penalized for bleak ending)- dialogue_density -- calibrate prose agent expectations- content_rating / reading_level -- adjust vocabulary/complexity expectations- story_epoch / time_units -- help consistency checker validate timeline### PrecedenceCLI flags > seed.yaml > defaults. If no seed.yaml exists and no CLI flags, run with defaults.### Implementation- Look for seed.yaml walking up from manuscript path- Parse the meta block only- Inject relevant fields into each agent's system prompt as author-stated-intent context- Extends the existing genre preamble pattern