feat(workflows): support video model artifacts

[DrHepa]

Closes #345

Summary

• add scalar video input support for model extension nodes
• add a durable Load Video workflow source and video artifact registration
• carry validated video snapshots through Electron, FastAPI, the extension subprocess, and the isolated runner
• preserve exact composite-model pinning, shared weights, weight variants, scene dispatch, existing video outputs/process nodes/multi-input declarations, and permissive third-party artifact kinds

Current base and review scope

This branch was rebuilt as one video-only commit on current dev at 221e44e9 after #348, #353, #357, #359, #360, #363, and #365 merged. It is no longer stacked: the scene-artifact, shared-weight, and weight-variant implementations now come directly from dev.

The new typed /generate/from-artifact execution path targets a scalar model input: "video". Existing video outputs, process-extension video I/O, and inputs arrays containing video remain valid and are not narrowed by this PR. Imported videos still reject traversal, symlink/reparse escapes, file replacement after validation, unsupported containers, and oversized files. No global artifact-type allowlist is introduced, so other third-party artifact kinds remain unaffected.

Validation

• Python suite: 233 passed, 2 skipped
• focused video/scene/workflow Node suites: 41 passed
• TypeScript --noEmit: passed
• ESLint: passed
• production build: passed
• git diff --check: passed
• full Node batch: one existing process-runner batch-interaction failure reproduced unchanged on clean current upstream/dev; the same test passes individually (3/3)
• prior consumer evidence: real local video to Pixal3D scene to WorldSculpt GLB generation passed