Add LLMCompress subsystem - sandbox for symbolic compression of LLM behavior

New subsystem fully self-contained:

Components:
- LLMCompress/llm_adapter.py: LLMAdapter + LLMResponse (abstract over LLM backends)
- LLMCompress/compression_report.py: CompressionReport (symbolic analysis results)
- LLMCompress/llm_compressor.py: compress_interaction() and compress_session()
- LLMCompress/tests/test_llm_compress.py: 5 comprehensive tests

Integration:
- Uses GlyphOS Cognitive Kernel for symbolic analysis
- Integrates with GlyphOS Event System
- Emits cognition.started and cognition.completed events
- Supports in-memory GX execution via execute_symbolic()

Test Coverage:
- LLMCompress tests: 5/5 PASS
- All existing tests still pass (52/52)
- Total: 57 tests passing

Bug fixes in cognitive_kernel.py:
- Fixed execute_symbolic() method calls to use correct function signatures
- normalize_segments(manifest, segments, payload)
- map_lanes(segments)
- build_envelope(manifest, lanes, payload, context)
- execute_with_lain(envelope)

Constraints:
- No modifications to gx_compiler/*
- No modifications to glyphs/super_registry.py
- Self-contained subsystem with proper isolation
- Full backward compatibility maintained

LLMCompress/tests/test_llm_compress.py

@@ -0,0 +1,133 @@
+"""Tests for LLMCompress subsystem.
+
+These tests verify:
+
+- LLMAdapter normalization
+- compress_interaction() end-to-end flow
+- compress_session() multi-turn flow
+- Integration with GlyphOS Cognitive Kernel
+- Event emission during compression
+
+This uses a fake LLM backend so tests run without external dependencies.
+"""
+
+from typing import Any, Dict
+
+from LLMCompress import (
+    LLMAdapter,
+    compress_interaction,
+    compress_session,
+    CompressionReport,
+)
+
+from glyphos.events import get_event_bus
+from glyphos.cognitive_kernel import get_kernel  # noqa: F401  # imported to ensure availability
+
+
+# ---------------------------------------------------------------------------
+# Fake LLM backend
+# ---------------------------------------------------------------------------
+
+def fake_llm_backend(prompt: str, **kwargs: Any) -> Dict[str, Any]:
+    """A deterministic fake LLM backend for testing."""
+    return {
+        "response": f"FAKE_RESPONSE({prompt})",
+        "tokens_prompt": len(prompt.split()),
+        "tokens_response": 3,
+        "model_name": "fake-llm-test",
+        "extra_info": "ok",
+    }
+
+
+# ---------------------------------------------------------------------------
+# Tests
+# ---------------------------------------------------------------------------
+
+def test_adapter_normalization():
+    adapter = LLMAdapter(fake_llm_backend, model_name="fake-llm-test")
+    out = adapter.run("hello world")
+
+    assert out.prompt == "hello world"
+    assert out.response.startswith("FAKE_RESPONSE")
+    assert out.tokens_prompt == 2
+    assert out.tokens_response == 3
+    assert out.model_name == "fake-llm-test"
+    assert isinstance(out.metadata, dict)
+
+
+def test_compress_interaction_basic():
+    adapter = LLMAdapter(fake_llm_backend, model_name="fake-llm-test")
+
+    bus = get_event_bus()
+    bus.clear_history()
+
+    report = compress_interaction(adapter, "hello test")
+
+    # Report structure
+    assert isinstance(report, CompressionReport)
+    assert len(report.interactions) == 1
+    assert "prompt" in report.interactions[0]
+    assert "response" in report.interactions[0]
+
+    # Kernel output
+    assert isinstance(report.fused_symbol, dict)
+    assert isinstance(report.diagnostics, dict)
+
+    # Events fired
+    history = bus.get_history(limit=10)
+    types = [e["type"] for e in history]
+
+    assert "cognition.started" in types
+    assert "cognition.completed" in types
+
+
+def test_compress_session_multi_turn():
+    adapter = LLMAdapter(fake_llm_backend, model_name="fake-llm-test")
+
+    bus = get_event_bus()
+    bus.clear_history()
+
+    prompts = ["turn one", "turn two", "turn three"]
+    report = compress_session(adapter, prompts)
+
+    assert isinstance(report, CompressionReport)
+    assert len(report.interactions) == 3
+
+    # Kernel output
+    assert isinstance(report.fused_symbol, dict)
+    assert isinstance(report.diagnostics, dict)
+
+    # Events fired
+    history = bus.get_history(limit=10)
+    types = [e["type"] for e in history]
+
+    assert "cognition.started" in types
+    assert "cognition.completed" in types
+
+
+def test_payload_encoding_is_valid_json():
+    adapter = LLMAdapter(fake_llm_backend)
+    report = compress_interaction(adapter, "encode me")
+
+    # Ensure payload was JSON-serializable and processed by kernel
+    assert isinstance(report.fused_symbol, dict)
+    assert isinstance(report.diagnostics, dict)
+
+
+def test_event_metadata_includes_source():
+    adapter = LLMAdapter(fake_llm_backend)
+
+    bus = get_event_bus()
+    bus.clear_history()
+
+    compress_interaction(adapter, "metadata test")
+
+    events = bus.get_history(limit=10)
+    found = False
+    for e in events:
+        if e["type"] == "cognition.started":
+            assert e["payload"]["source"] == "LLMCompress"
+            found = True
+            break
+
+    assert found, "Expected cognition.started event with source=LLMCompress"