bug: confidenceRange returns inverted sentinels when no effects have classification

[yvonnedevlinrh]

Summary

confidenceRange in internal/adapter/contract.go (lines 194-203) initializes sentinels as minConf = 100, maxConf = 0 and skips effects with nil Classification. When all effects have nil Classification (e.g., an external analyzer that doesn't support classify_signals), the sentinels are returned unchanged, producing an inverted range where min (100) > max (0).

This value flows through deriveCoverageReason (line 186-187) into ContractCoverageInfo.MinConfidence and MaxConfidence, potentially surfacing in output as "confidence 100-0".

Impact

Low. Only triggers when an external analyzer returns side effects without classification data. The output is confusing but does not affect CRAP score computation or cause a crash.

Reproduction

Use an external analyzer whose initialize response sets capabilities.classify_signals: false. Run gaze crap --analyzer <binary> against a package where all detected effects lack classification. The confidence range in the output will show 100-0.

Suggested Fix

Return (0, 0) when no effects have classifications:

func confidenceRange(effects []taxonomy.SideEffect) (minConf, maxConf int) {
	minConf, maxConf = 100, 0
	found := false
	for _, e := range effects {
		if e.Classification == nil {
			continue
		}
		found = true
		minConf = min(minConf, e.Classification.Confidence)
		maxConf = max(maxConf, e.Classification.Confidence)
	}
	if !found {
		return 0, 0
	}
	return minConf, maxConf
}

## References
- Introduced in PR #178 (commit 131cd1f, refined in 711bbb0)
- File: internal/adapter/contract.go:194-203
- Caller: deriveCoverageReason at contract.go:186

Found-by: OpenCode (claude-opus-4-6)