feat: comprehensive token usage tracking for V2

Add provider-specific usage details, fix streaming usage, and return
usage from all high-level APIs (Chat.Send, Generate[T], Agent.Run).

Breaking changes:
- Chat.Send/SendMessage/SendWithImages now return (string, *Usage, error)
- Generate[T]/GenerateWith[T] now return (T, *Usage, error)
- Agent.Run/RunMessages now return (string, *Usage, error)

New features:
- Usage.Details map for provider-specific token breakdowns
  (reasoning, cached, audio, thoughts tokens)
- OpenAI streaming now captures usage via StreamOptions.IncludeUsage
- Google streaming now captures UsageMetadata from final chunk
- UsageTracker.Details() for accumulated detail totals
- ModelPricing and PricingRegistry for cost computation

Closes #2

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

v2/agent/agent.go

@@ -18,7 +18,7 @@
 //	        coder.AsTool("code", "Write and run code"),
 //	    )),
 //	)
-//	result, err := orchestrator.Run(ctx, "Build a fibonacci function in Go")
+//	result, _, err := orchestrator.Run(ctx, "Build a fibonacci function in Go")
 package agent
 
 import (
@@ -64,13 +64,15 @@ func New(model *llm.Model, system string, opts ...Option) *Agent {
 
 // Run executes the agent with a user prompt. Each call is a fresh conversation.
 // The agent loops tool calls automatically until it produces a text response.
-func (a *Agent) Run(ctx context.Context, prompt string) (string, error) {
+// Returns the text response, accumulated token usage, and any error.
+func (a *Agent) Run(ctx context.Context, prompt string) (string, *llm.Usage, error) {
 	return a.RunMessages(ctx, []llm.Message{llm.UserMessage(prompt)})
 }
 
 // RunMessages executes the agent with full message control.
 // Each call is a fresh conversation. The agent loops tool calls automatically.
-func (a *Agent) RunMessages(ctx context.Context, messages []llm.Message) (string, error) {
+// Returns the text response, accumulated token usage, and any error.
+func (a *Agent) RunMessages(ctx context.Context, messages []llm.Message) (string, *llm.Usage, error) {
 	chat := llm.NewChat(a.model, a.reqOpts...)
 	if a.system != "" {
 		chat.SetSystem(a.system)
@@ -107,7 +109,8 @@ type delegateParams struct {
 func (a *Agent) AsTool(name, description string) llm.Tool {
 	return llm.Define[delegateParams](name, description,
 		func(ctx context.Context, p delegateParams) (string, error) {
-			return a.Run(ctx, p.Input)
+			text, _, err := a.Run(ctx, p.Input)
+			return text, err
 		},
 	)
 }