Refactor framework core for planned dense/compact execution

[Concode0]

Large-scale framework refactor that makes Versor’s core architecture the base for scalable model development.

Includes:

• New core foundation/planning/runtime structure
• Dense and compact grade-layout execution paths
• Pairwise projected products for mixed compact lane widths
• Product layers and functional product helpers
• Centralized manifold tagging and optimizer factory support
• Framework pipeline tests for dense, compact, pairwise, and optimizer-connected execution
• 8D framework pipeline benchmark
• Architecture documentation with Mermaid diagrams
• Python support floor raised to 3.10

Validations:

• 8D benchmark: compact planned paths are much faster than dense full GP
• Fullgraph compile: 8D compact layer pipeline captures as one graph with no graph breaks, and also runs with inductor fullgraph.