feat: conversion-driven extensions — resolvers, DefineTool, hooks, ops controls

Phase 9a (ADR-0014): Registry.RegisterResolver for dynamic tiers;
DefineTool[Args] typed tools; Usage cache/reasoning detail fields wired
through anthropic/openai/google; WithPromptCaching (Anthropic
cache_control); agent supervision hooks (WithMaxStepsFunc, WithSteer,
WithCompactor, WithToolErrorLimits + ErrToolLoop); health
Bench/Unbench/Snapshot; ChainConfig.Observer failover events.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

provider/anthropic/wire.go

@@ -24,6 +24,13 @@ type wireRequest struct {
 	TopP          *float64          `json:"top_p,omitempty"`
 	StopSequences []string          `json:"stop_sequences,omitempty"`
 	OutputConfig  *wireOutputConfig `json:"output_config,omitempty"`
+	// CacheControl is the top-level auto-placement form of prompt caching:
+	// the API puts the breakpoint on the last cacheable block.
+	CacheControl *wireCacheControl `json:"cache_control,omitempty"`
+}
+
+type wireCacheControl struct {
+	Type string `json:"type"`
 }
 
 type wireMessage struct {
@@ -109,8 +116,10 @@ type wireUsage struct {
 // real total input is input + cache_creation + cache_read.
 func (u wireUsage) toUsage() llm.Usage {
 	return llm.Usage{
-		InputTokens:  u.InputTokens + u.CacheCreationInputTokens + u.CacheReadInputTokens,
-		OutputTokens: u.OutputTokens,
+		InputTokens:      u.InputTokens + u.CacheCreationInputTokens + u.CacheReadInputTokens,
+		OutputTokens:     u.OutputTokens,
+		CacheReadTokens:  u.CacheReadInputTokens,
+		CacheWriteTokens: u.CacheCreationInputTokens,
 	}
 }
 
@@ -157,6 +166,11 @@ func buildWireRequest(modelID string, req llm.Request, defaultMax int, stream bo
 			Schema: req.Schema,
 		}}
 	}
+	if req.PromptCache {
+		// Top-level auto-placement: the API puts the cache breakpoint on
+		// the last cacheable block.
+		wr.CacheControl = &wireCacheControl{Type: "ephemeral"}
+	}
 	return wr
 }