Fix read tool token limit - reduce to 20k for MCP overhead

- Reduced max_tokens from 25000 → 20000
- Prevents 'response exceeds maximum allowed tokens' errors
- Leaves buffer for MCP protocol overhead

Fixes: read tool returning 27444 tokens (exceeds 25000 limit)

Author: hanzo-dev

pkg/hanzo-mcp/hanzo_mcp/tools/filesystem/read.py

@@ -212,9 +212,10 @@ async def call(
                 await tool_ctx.info(f"Successfully read file: {file_path}")
 
                 # Apply token limit to prevent excessive output
+                # Reduce to 20000 to leave buffer for MCP overhead
                 return truncate_response(
                     result,
-                    max_tokens=25000,
+                    max_tokens=20000,
                     truncation_message="\n\n[File content truncated due to token limit. Use offset/limit parameters to read specific sections.]",
                 )