feat(pipeline): introduce binary-safe artifact kernel

[szmyty]

Summary

Introduces the binary-safe artifact substrate tracked by #352 without prematurely stabilizing the Transform v2/plugin contract owned by #357.

Artifact kernel

• adds first-class Artifact, ArtifactId, SHA-256 ArtifactDigest, media type, canonical format, byte size, metadata, ordered lineage, payload handle, and storage-class types
• separates content-addressed payload identity from artifact-record identity so payload bytes can be deduplicated without collapsing distinct format/lineage records
• adds ordered ArtifactCollection values for aggregation inputs
• adds a file-backed ArtifactStore with streaming SHA-256 ingestion and deterministic object paths
• keeps canonical source files immutable by importing them into the store
• atomically materializes final outputs only after a complete copy and sync

Artifact-native DAG execution

• makes artifact values the canonical DagExecutor substrate
• adds execute_artifact(...) and execute_artifacts(...)
• adds register_artifact(...) for binary-safe transforms
• preserves the existing execute(..., String) and register_single(..., Transform) APIs through TextTransformAdapter
• rejects binary payloads explicitly at the legacy UTF-8 boundary rather than decoding them lossily
• passes collection inputs to aggregation transforms as ordered file-backed payload paths
• imports aggregation output into the content-addressed store only after the transform succeeds

Cache

• adds an artifact-native DAG cache
• node keys use artifact-record identity, SHA-256 digest, byte size, source/target formats, and transform cache identity instead of re-hashing complete in-memory strings
• validates cached payload presence before reuse
• treats legacy string-cache files as cache misses during migration
• leaves complete provider/tool/environment fingerprinting to Stabilize Transform v2 and the plugin SDK around typed artifact I/O #357/Create a runtime tool capability registry and reproducible toolchain fingerprint #359 rather than inventing unstable semantics here

Graph build integration

• removes fs::read_to_string from graph source ingestion
• imports configured input files as binary-safe source artifacts
• keeps work/cache artifacts in hidden Renderflow state rather than emitting intermediates as final outputs
• atomically materializes selected graph outputs
• logs artifact ID, digest, and byte size

Fixtures and documentation

Tests cover:

• representative PDF, PNG, WAV, and Markdown artifact payloads
• binary graph traversal containing invalid UTF-8 bytes
• legacy text-transform compatibility
• ordered collection aggregation and lineage
• artifact-cache reuse
• failed aggregation leaving no falsely complete content-addressed artifact
• source immutability and atomic final replacement

docs/artifact-kernel.md documents the model, store lifecycle, cache identity, compatibility boundary, Flow projection boundary, and migration sequence.

The PR also repairs the supported-format documentation generator for the Rust workspace layout and commits its deterministic regenerated projection. This surfaced and corrected previously stale graph identifiers without hand-editing generated documentation.

Architecture boundaries

This PR intentionally does not:

• claim production transform support for every Format variant
• replace the public legacy Transform trait or plugin SDK (Stabilize Transform v2 and the plugin SDK around typed artifact I/O #357)
• centralize process execution policy (Centralize and harden external tool execution with bounded process policies #356)
• define complete toolchain fingerprints (Create a runtime tool capability registry and reproducible toolchain fingerprint #359)
• make execution results provenance-complete (Make execution results provenance-complete and align artifact evidence with Flow contracts #355)
• couple Renderflow core to Flow (Add checkpointed execution and a resumable Renderflow provider seam for Flow #358)

Validation

Final head: 1dc01301576600110fa8df5ae3e59c07e2dfeba9

All final-head workflows are green:

• CI run 33429687003
  • Clippy with warnings denied
  • Rust workspace build and full tests
  • web frozen install, format, lint, typecheck, tests, build, and artifact upload
  • AUR stable and git package validation
  • portable installer smoke test
  • Snap build, classic install, renderflow --version smoke test, and artifact upload
  • bundled dry-run, library embedding, and plugin registration examples
• Docs run 33429687002
  • generated-doc regeneration
  • zero-diff generated-doc verification
  • MkDocs build
  • versioned docs smoke build
• Commit Lint run 33429687036

The PR contains one Conventional Commit and is mergeable.

Closes #352

[devactivity-app]

Pull Request Summary by devActivity

Metrics

Achievements

• @szmyty
  Earned XP: 15⭐
  Check Your Progress