fix(deepseek): add more exhaustive sanitation for deepseek-reasoner

R1 model does not support presence_penalty, frequency_penalty, logit_bias,
logprobs, or top_logprobs. Added these to the list of stripped parameters.

src/providers/deepseek.rs

@@ -66,6 +66,11 @@ impl super::Provider for DeepSeekProvider {
                 obj.remove("tool_choice");
                 obj.remove("temperature");
                 obj.remove("top_p");
+                obj.remove("presence_penalty");
+                obj.remove("frequency_penalty");
+                obj.remove("logit_bias");
+                obj.remove("logprobs");
+                obj.remove("top_logprobs");
             }
         }
 
@@ -129,11 +134,16 @@ impl super::Provider for DeepSeekProvider {
         if request.model == "deepseek-reasoner" {
             if let Some(obj) = body.as_object_mut() {
                 obj.remove("stream_options");
-                // Also does not support tools/temperature/top_p
+                // Also does not support these parameters
                 obj.remove("tools");
                 obj.remove("tool_choice");
                 obj.remove("temperature");
                 obj.remove("top_p");
+                obj.remove("presence_penalty");
+                obj.remove("frequency_penalty");
+                obj.remove("logit_bias");
+                obj.remove("logprobs");
+                obj.remove("top_logprobs");
             }
         } else {
             // For standard deepseek-chat, keep it clean