Update LLM integration and add new agent tools and utilities

Refactored LLM handling to use updated langchaingo models and tools, replacing gollm dependencies. Introduced agent-related utilities, tools, and counters for better modular functionality. Added a parser for LLM model configuration and revamped the answering mechanism with enhanced support for tool-based interaction.

pkg/agent/ask.go

@@ -0,0 +1,46 @@
+package agent
+
+import (
+	"encoding/json"
+	"fmt"
+)
+
+var AskTool = FromFunction(
+	func(ctx *Context, args struct {
+		Question string `description:"the question to answer"`
+	}) (FuncResponse, error) {
+		var q Question
+
+		q.Question = args.Question
+		ctx = ctx.WithQuestion(q)
+
+		answers, err := ask(ctx, q)
+
+		if err != nil {
+			return FuncResponse{}, err
+		}
+
+		tb := ToolBox{}
+		tb.Register(SummarizeAnswers)
+
+		b, err := json.Marshal(answers.Answers)
+		if err != nil {
+			return FuncResponse{}, fmt.Errorf("failed to marshal answers: %w", err)
+		}
+
+		q = Question{Question: string(b)}
+		ctx = ctx.WithQuestion(q)
+		answers, err = tb.Run(ctx, q)
+
+		if err != nil {
+			return FuncResponse{}, fmt.Errorf("failed to summarize answers: %w", err)
+		}
+
+		if len(answers.Answers) == 0 {
+			return FuncResponse{}, fmt.Errorf("no response from model")
+		}
+
+		return FuncResponse{Result: answers.Answers[0].Answer}, nil
+	}).
+	WithName("ask").
+	WithDescription("Ask the agent a question, this is useful for splitting a question into multiple parts")