Add memba MVP: C++ core, Python SDK, CLI, examples, experiments

C++ core (libmemba.so):
- include/memba/state.h — C API (state_new/free/save/load/get_size)
- src/state.cpp — MEMB file format: magic, version, SHA-256 model_id,
  CRC-32, opaque llama_state_*_data() blob
- src/cli.cpp — minimal demo binary with greedy sampler
- CMakeLists.txt + build.sh with llama.cpp submodule, CUDA auto-detect

Python SDK (memba):
- core.py — file I/O via llama-cpp-python's exposed C functions,
  unwraps _LlamaContext to access raw context pointer (≥0.3.x)
- session.py — high-level Session with auto-save/load, ChatML wrapper
  for instruct models, raw mode for base models
- cli.py — typer-based: chat (REPL), run (one-shot), list, rm, info

Examples:
- 01_basic_save_load.py, 02_chat_session.py

Experiments (throwaway POCs documenting product-direction findings):
- recall_poc.py — git log → state → cross-process query
- mood_poc.py — batch sentiment trajectory, Mamba vs Transformer
- mood_stream_poc.py, mood_batch_poc.py — variants
- diag_saveload.py — minimal save/load isolation test
- README.md documents the headline finding: save/load is byte-identical,
  but Falcon-Mamba-7B-Instruct does not retain facts across conversation
  turns even in-process — limits viable products to single-prompt analysis
  and persona priming.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

experiments/diag_saveload.py

@@ -0,0 +1,57 @@
+"""
+Diagnostic: does the model retain context AFTER save in the SAME process,
+and AFTER load in a fresh process? Compares three scenarios.
+"""
+import sys, argparse
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "python"))
+from llama_cpp import Llama
+from memba import core
+
+MODEL = "/home/emil/Desktop/Coding/AI/Memba/falcon-mamba-7B-instruct-Q4_K_M.gguf"
+STATE = "/tmp/diag_saveload.memb"
+
+INGEST  = "I'm telling you a secret. My pet hamster is named Bartholomew. He is 4 years old. Reply 'ok'."
+QUERY   = "What is the name of my pet?"
+
+
+def chatml(msg):
+    return f"<|im_start|>user\n{msg}<|im_end|>\n<|im_start|>assistant\n"
+
+
+def make_llama():
+    return Llama(model_path=MODEL, n_ctx=2048, n_gpu_layers=-1, verbose=False)
+
+
+def ask(m, prompt):
+    out = m(chatml(prompt), max_tokens=60, stop=["<|im_end|>"], echo=False)
+    return out["choices"][0]["text"].strip()
+
+
+def build():
+    m = make_llama()
+    ack = ask(m, INGEST)
+    print(f"  [build] ack: {ack!r}")
+    print(f"  [build] state size (live): {core.get_state_size(m):,} B")
+    # In-process query BEFORE saving
+    print(f"  [build] in-proc query BEFORE save: {ask(m, QUERY)!r}")
+    # Save
+    core.save_state(m, MODEL, STATE)
+    print(f"  [build] state saved")
+    # In-process query AFTER saving (should still work — save shouldn't mutate)
+    print(f"  [build] in-proc query AFTER save:  {ask(m, QUERY)!r}")
+
+
+def query():
+    m = make_llama()
+    print(f"  [query] before load — fresh model: {ask(m, QUERY)!r}")
+    core.load_state(m, MODEL, STATE)
+    print(f"  [query] state size after load: {core.get_state_size(m):,} B")
+    print(f"  [query] after load:                {ask(m, QUERY)!r}")
+    # Try a second time in case position is wonky
+    print(f"  [query] second ask:                {ask(m, QUERY)!r}")
+
+
+if __name__ == "__main__":
+    cmd = sys.argv[1] if len(sys.argv) > 1 else "build"
+    {"build": build, "query": query}[cmd]()