v0.1.2

• Remove architecture-specific IGNORE list adapted from llama.cpp
• Use (\d+) wildcard for uniform layers, grouped alternation for mixed layers
• Detect and raise on conflicting tensor qtypes across template GGUFs