wam: reject non-positive input weights instead of returning NaN

[arpitjain099]

What

WeightedArithmeticMean.Score computes itemSum / itemCount, where itemCount is the running sum of the matched input weights. When that total weight is zero, the result is math.NaN(). Because Score returns a bare float64, it has no way to signal the failure, so an undefined score propagates silently to the caller.

This validates the weights up front in wam.New instead. Any input with a zero or negative weight is rejected with an error:

• A zero weight makes the total weight zero, so the weighted mean is undefined (the divide-by-zero that produces NaN).
• A negative weight has no meaningful interpretation in a weighted average.

This follows @calebbrown's direction on the issue: "Non-positive weights don't make much sense, and should cause an error if supplied." New already returns (algorithm.Algorithm, error), so this is the natural place to surface the problem, and Score's behavior for valid inputs is unchanged.

Why this is safe for existing configs

All shipped scorer configs (config/scorer/*.yml) use only positive weights (1, 2, 0.5), so valid configurations are unaffected. The new error only fires on a misconfigured zero or negative weight.

Tests

Extended the existing table-driven test in wam_test.go:

• zero weight -> New returns an error (previously Score returned NaN)
• negative weight -> New returns an error (previously Score silently returned a wrong value)
• a single positive weight -> accepted, correct score

Before this change those first two cases produce NaN / a silently-wrong value; after, they return a clear error. go test ./internal/scorer/... passes, and gofmt/go vet are clean.

Fixes #314