feat: add FOREMAN_KEEP_ALIVE config for worker model residency

Allow configuring how long the worker model stays resident on the Ollama
target after a request via FOREMAN_KEEP_ALIVE env var. Accepts Ollama
duration strings ("-1" forever, "0" unload, "15m", "1h", etc). Defaults
to "-1" (pin forever). The embedder warm-up is unaffected and always
uses keep_alive=-1.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

internal/config/config.go

@@ -48,6 +48,13 @@ type Config struct {
 	// JobTTL is how long terminal jobs are retained before the pruner deletes them
 	// (default 24h).
 	JobTTL time.Duration
+
+	// KeepAlive is the keep_alive value sent in outbound /api/chat requests to the
+	// Ollama target. It controls how long the worker model stays resident after a
+	// request. Accepts Ollama duration strings like "15m", "1h", "-1" (forever), or
+	// "0" (unload immediately). Default is "-1" (pin forever). This does NOT affect
+	// the embedder, which is always pinned with keep_alive=-1.
+	KeepAlive string
 }
 
 // Load reads configuration from environment variables and returns a validated Config.
@@ -64,6 +71,7 @@ func Load() (Config, error) {
 		EmbedModel:    os.Getenv("FOREMAN_EMBED_MODEL"),
 		DBPath:        envOr("FOREMAN_DB_PATH", "foreman.db"),
 		WebhookSecret: os.Getenv("FOREMAN_WEBHOOK_SECRET"),
+		KeepAlive:     envOr("FOREMAN_KEEP_ALIVE", "-1"),
 	}
 
 	pollStr := envOr("FOREMAN_POLL_INTERVAL", "30s")