feat: enhance error handling and reasoning mechanisms across middleware

- Improve error handling across multiple middleware and adapter components, ensuring consistent error response formats in JSON.
- Enhance the functionality of request conversion functions by including context parameters and robust error wrapping.
- Introduce new features related to reasoning content in the messaging model, providing better customization and explanations in the documentation.

relay/adaptor/aws/claude/adapter.go

@@ -21,7 +21,11 @@ func (a *Adaptor) ConvertRequest(c *gin.Context, relayMode int, request *model.G
 		return nil, errors.New("request is nil")
 	}
 
-	claudeReq := anthropic.ConvertRequest(*request)
+	claudeReq, err := anthropic.ConvertRequest(c, *request)
+	if err != nil {
+		return nil, errors.Wrap(err, "convert request")
+	}
+
 	c.Set(ctxkey.RequestModel, request.Model)
 	c.Set(ctxkey.ConvertedRequest, claudeReq)
 	return claudeReq, nil

relay/adaptor/aws/claude/main.go

@@ -88,7 +88,7 @@ func Handler(c *gin.Context, awsCli *bedrockruntime.Client, modelName string) (*
 		return utils.WrapErr(errors.Wrap(err, "unmarshal response")), nil
 	}
 
-	openaiResp := anthropic.ResponseClaude2OpenAI(claudeResponse)
+	openaiResp := anthropic.ResponseClaude2OpenAI(c, claudeResponse)
 	openaiResp.Model = modelName
 	usage := relaymodel.Usage{
 		PromptTokens:     claudeResponse.Usage.InputTokens,
@@ -159,7 +159,7 @@ func StreamHandler(c *gin.Context, awsCli *bedrockruntime.Client) (*relaymodel.E
 				return false
 			}
 
-			response, meta := anthropic.StreamResponseClaude2OpenAI(claudeResp)
+			response, meta := anthropic.StreamResponseClaude2OpenAI(c, claudeResp)
 			if meta != nil {
 				usage.PromptTokens += meta.Usage.InputTokens
 				usage.CompletionTokens += meta.Usage.OutputTokens