Implement OpenCog cognitive architecture with tensor-based embeddings and PLN reasoning

[Copilot]

For changes to the core ggml library (including to the CMake build system), please open a PR in https://github.com/ggml-org/llama.cpp. Doing so will make your PR more visible, better tested and more likely to be reviewed.

Description

Implements OpenCog cognitive architecture using GGML tensor operations for semantic reasoning and attention allocation.

Core Features:

• Tensor embeddings: Vector representations for atoms with compositional semantics (links blend target embeddings)
• PLN reasoning: Deduction, induction, abduction, revision, modus ponens with probabilistic truth values
• ECAN attention: STI/LTI/VLTI allocation with automatic attention spreading through graph
• Pattern matching: Cosine similarity-based matching on embeddings with sorted results
• Compute similarity: Fast embedding comparisons for semantic reasoning

Example usage:

auto* atomspace = ggml_opencog_atomspace_new(128);

// Create atoms with embeddings
uint64_t dog_id = ggml_opencog_add_atom(atomspace, GGML_OPENCOG_CONCEPT_NODE, 
                                        "Dog", {0.9f, 0.8f}, {});
uint64_t mammal_id = ggml_opencog_add_atom(atomspace, GGML_OPENCOG_CONCEPT_NODE,
                                           "Mammal", {0.9f, 0.8f}, {});

// Create link (automatically blends embeddings)
ggml_opencog_add_atom(atomspace, GGML_OPENCOG_INHERITANCE_LINK,
                     "Dog->Mammal", {0.95f, 0.9f}, {dog_id, mammal_id});

// Compute semantic similarity
float sim = ggml_opencog_compute_similarity(atomspace, dog_id, mammal_id);

// Allocate attention
ggml_opencog_update_attention(atomspace, dog_id, 15.0f, 2.0f);

Implementation details:

• Embeddings stored as std::vector<float> to avoid per-atom tensor context overhead
• Type embeddings shared across atoms, combined with hash-based perturbation
• PLN formulas follow standard confidence propagation rules
• Attention spreads to connected atoms at 10% factor with [-100, 100] STI clamping

Testing:

• 9 unit tests covering all features (embeddings, similarity, attention, PLN operations)
• 3 example programs: basic, reasoning, advanced features
• Comprehensive README documentation

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.