Fix false positives in agent guardrails

Skip empty fingerprints in SystemPromptLeakGuardrail to prevent blank
values from flagging every message. Replace overly broad "sk-" pattern
in SensitiveDataGuardrail with specific LLM provider prefixes
("sk-proj-" for OpenAI, "sk-ant-" for Anthropic) to avoid false
positives on common words like "risk-based" or "task-management".

Signed-off-by: Bryan Frimin <bryan@getprobo.com>

pkg/agent/guardrail/system_prompt_leak.go

@@ -27,9 +27,12 @@ type SystemPromptLeakGuardrail struct {
 }
 
 func NewSystemPromptLeakGuardrail(fingerprints []string) *SystemPromptLeakGuardrail {
-	lowered := make([]string, len(fingerprints))
-	for i, f := range fingerprints {
-		lowered[i] = strings.ToLower(f)
+	lowered := make([]string, 0, len(fingerprints))
+	for _, f := range fingerprints {
+		if f == "" {
+			continue
+		}
+		lowered = append(lowered, strings.ToLower(f))
 	}
 
 	return &SystemPromptLeakGuardrail{fingerprints: lowered}