Greptile Summary

This is a documentation-only PR that consolidates seed reader documentation for filesystem and agent rollout sources, adds a new FileSystemSeedReader plugin authoring guide, and introduces a runnable Markdown Section Seed Reader recipe. It also fixes a real bug where designer.preview() was incorrectly used in the Complete Example instead of data_designer.preview().

Key changes:

• docs/concepts/seed-datasets.md: Adds DirectorySeedSource, FileContentsSeedSource, and AgentRolloutSeedSource sections with column listings, code examples, and cross-links. Fixes the designer.preview() → data_designer.preview() bug in the Complete Example section.
• docs/plugins/filesystem_seed_reader.md (new): Step-by-step guide covering the build_manifest/hydrate_row contract, inline reader pattern (no packaging required), manifest-based selection semantics, and the optional packaging path. Accurately reflects the engine implementation.
• docs/assets/recipes/plugin_development/markdown_seed_reader.py (new): Self-contained, runnable uv run recipe that fans each Markdown file into one row per ATX heading section. Logic, output_columns schema, and manifest-based IndexRange selection all check out correctly against the framework source.
• docs/recipes/plugin_development/markdown_seed_reader.md (new): Landing page for the recipe with correct relative paths, --8<-- snippet include, and download link.
• docs/plugins/overview.md, docs/plugins/example.md, docs/recipes/cards.md, mkdocs.yml: Navigation and cross-link updates to surface the new content.

Important Files Changed

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User calls DataDesigner.preview / create] --> B{Seed source type?}
    B -->|LocalFile / HuggingFace / DataFrame| C[Built-in SeedReader]
    B -->|DirectorySeedSource / FileContentsSeedSource| D[FileSystemSeedReader]
    B -->|AgentRolloutSeedSource| E[AgentRolloutSeedReader]
    B -->|DirectorySeedSource + custom reader| F[Custom FileSystemSeedReader\ne.g. MarkdownSectionDirectorySeedReader]
    D --> G[build_manifest\nenumerate matching files]
    F --> G
    G --> H[IndexRange / PartitionBlock / shuffle\noperates on manifest rows]
    H --> I[hydrate_row per manifest row\n1:1 or 1:N fanout]
    I --> J[output_columns schema validation]
    J --> K[DuckDB registration\n→ seed columns available in Jinja2 templates]
    E --> K
    C --> K

_{Reviews (1): Last reviewed commit: "docs: consolidated seed reader documenta..." | Re-trigger Greptile}