implement aicpp: C/C++ cognitive kernel on GGML/llama.cpp

[Copilot]

Adds a complete C/C++ implementation of AIChat as a cognitive kernel system (cpp/) built on GGML and llama.cpp, following the Echo.Kern/OpenCog architecture.

Architecture

• Kernel layer (C11) — bootstrap (4-stage), membrane-aware scheduler, tensor-backed allocator (aligned_alloc), hypergraph FS, mutex/spinlock/semaphore, nanosecond timers
• Cognitive layer (C++) — AtomSpace (hypergraph), ECAN attention allocation, PLN inference, ESN reservoir computing, CognitiveLoop orchestrating all four
• LLM layer (C++) — llama.cpp model loading, chat completion with streaming via sampler chain
• CLI layer (C++) — arg parser, readline REPL with /clear//history commands, JSON session persistence
• RAG — hash-projection (FNV-1a, 1024-dim) vector store, cosine-similarity retrieval, no external deps
• Function calling — <tool_call>/<tool_result> protocol, callback + shell-command tool types

Bug fixes

• CMakeLists.txt: CMAKE_C_STANDARD 99 → 11; aligned_alloc is C11-only, was generating implicit-declaration warnings
• bootstrap.c: added #include "aichat/cognitive.h" to resolve implicit declarations of atomspace_init, ecan_init, pln_init (defined in C++ TUs)

[drzo]

coool