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>

llm/request.go

@@ -43,6 +43,11 @@ type Request struct {
 	// Providers map it to their native knob (OpenAI reasoning_effort,
 	// Ollama think levels) and ignore it where no mapping exists.
 	ReasoningEffort string
+
+	// PromptCache opts the request into the provider's prompt caching
+	// (Anthropic cache_control; ignored by providers that cache
+	// automatically or not at all).
+	PromptCache bool
 }
 
 // Option mutates a Request before it is sent. Options passed to Generate or
@@ -100,6 +105,12 @@ func WithReasoningEffort(level string) Option {
 	return func(r *Request) { r.ReasoningEffort = level }
 }
 
+// WithPromptCaching opts into provider prompt caching where it is an
+// explicit feature (Anthropic); a no-op elsewhere.
+func WithPromptCaching() Option {
+	return func(r *Request) { r.PromptCache = true }
+}
+
 // Apply returns a copy of the request with all options applied. Providers
 // and wrappers call this once at the top of Generate/Stream.
 func (r Request) Apply(opts ...Option) Request {