Use json v2

internal/llm/client_completions.go

@@ -4,7 +4,7 @@ import (
 	"bufio"
 	"bytes"
 	"context"
-	"encoding/json"
+	"encoding/json/v2"
 	"fmt"
 	"io"
 	"log/slog"
@@ -83,7 +83,7 @@ func (c *Client) completions(
 		ThinkingBudgetTokens: p.ThinkingBudgetTokens,
 		ChatTemplateKWARGS:   kwargs,
 		ParallelToolCalls:    new(true),
-	})
+	}, json.FormatNilSliceAsNull(true))
 	if err != nil {
 		return nil, fmt.Errorf(
 			"completions: marshal request body: %w", err,
@@ -152,7 +152,7 @@ func (c *Client) completions(
 	}
 
 	var cr = &ChatResponse{}
-	if err := json.NewDecoder(res.Body).Decode(cr); err != nil {
+	if err := json.UnmarshalRead(res.Body, cr); err != nil {
 		return nil, fmt.Errorf("completions: decode response: %w", err)
 	}
 	if len(cr.Choices) == 0 {
@@ -283,7 +283,7 @@ func (c *Client) CompletionsStream(
 		StreamOptions:        &StreamOptions{IncludeUsage: true},
 		ThinkingBudgetTokens: p.ThinkingBudgetTokens,
 		Tools:                p.Tools,
-	})
+	}, json.FormatNilSliceAsNull(true))
 	if err != nil {
 		cancel()
 		return nil, fmt.Errorf(