feat: M1-M7 gap audit phase 3 — context prefix, deferred tools, compact hooks

Gap 11 (M6): Fixed context prefix
- Window.PrefixMessages stores immutable docs (CLAUDE.md, .gnoma/GNOMA.md)
- Prefix stripped before compaction, prepended after — survives all compaction
- AllMessages() returns prefix + history for provider requests
- main.go loads CLAUDE.md and .gnoma/GNOMA.md at startup as prefix

Gap 12 (M6): Deferred tool loading
- DeferrableTool optional interface: ShouldDefer() bool
- buildRequest() skips deferred tools until activated
- Tools auto-activate on first model request (activatedTools map)
- agent + spawn_elfs marked as deferrable (large schemas, rarely needed early)
- Saves ~800 tokens per deferred tool per request

Gap 13 (M6): Pre/post compact hooks
- OnPreCompact/OnPostCompact callbacks in WindowConfig
- Called in doCompact() (shared by CompactIfNeeded + ForceCompact)
- M8 hooks system will extend these to full protocol

internal/engine/engine.go

@@ -55,6 +55,10 @@ type Engine struct {
 	// Cached model capabilities, resolved lazily
 	modelCaps    *provider.Capabilities
 	modelCapsFor string // model ID the cached caps are for
+
+	// Deferred tool loading: tools with ShouldDefer() are excluded until
+	// the model requests them. Activated on first use.
+	activatedTools map[string]bool
 }
 
 // New creates an engine.
@@ -67,8 +71,9 @@ func New(cfg Config) (*Engine, error) {
 		logger = slog.Default()
 	}
 	return &Engine{
-		cfg:    cfg,
-		logger: logger,
+		cfg:            cfg,
+		logger:         logger,
+		activatedTools: make(map[string]bool),
 	}, nil
 }