feat: initial llm-connect package scaffold

Copy markitect.llm module into standalone llm_connect package.
All markitect.* imports replaced with llm_connect.* equivalents.
LLMError base class inlined (no markitect.exceptions dependency).
Verified: from llm_connect import create_adapter works.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

llm_connect/_token_estimator.py

@@ -0,0 +1,16 @@
+"""
+Rough token estimation for backends that don't return usage data.
+
+Uses the ~4 characters per token heuristic common across English LLM tokenizers.
+"""
+
+
+def estimate_tokens(text: str) -> int:
+    """Estimate the number of tokens in *text*.
+
+    This is intentionally coarse — it is only used by the Claude Code CLI
+    adapter where real token counts are unavailable.
+    """
+    if not text:
+        return 0
+    return max(1, len(text) // 4)