feat: add configurable commercial scoring weights

[yigitcan-ozturk]

What changed

• add strict configurable scoring profiles for price, lead time and payment terms
• add --weights CLI support using auditable JSON configuration files
• preserve the current default 50/30/20 scoring model when no profile is supplied
• validate required criteria, reject unsupported criteria and require weights to sum to 1.0
• return the effective scoring profile in the machine-readable payload
• expose weight loading/validation helpers through the public Python API
• add tests proving custom profiles can change supplier recommendations
• add a sample weight profile and update the roadmap/documentation

Compatibility

• existing CLI commands continue to use the default weights
• quotation input formats and the v0.2 JSON structure remain compatible
• custom criteria remain deliberately unsupported to preserve the tool's commercial decision boundary

Why

This makes rfqdiff adaptable to different procurement strategies without hiding decision logic or weakening auditability.