hardening: normalize fusion weights and bound combined scores

[Djain912]

Summary

Hardens combined fusion behavior by normalizing unsafe weight inputs and bounding merged scores.

Changes

• Normalize and sanitize lexical/embedding fusion weights.
• Add safe defaults when provided weights are invalid.
• Bound combined scores into [0,1].
• Add robust regression tests around weight semantics and score bounds.

Validation

• go test ./...

Stack Flow

• PR 32 (base hardening): hardening: normalize fusion weights and bound combined scores #32
• PR 33 (runtime cancellation hardening): hardening: add cancellation-safe matcher execution paths #33
• PR 34 (adversarial options matrix): test: add adversarial FindOptions hardening matrix #34

[luigiagent]

Thanks for this work, it’s still useful. We’ve opened #39 as a superseding PR that carries the same hardening direction forward on top of the newer stacked branch history, so review/merge should continue there instead. We’ll use #39 as the replacement path for this change set.