feat(launcher): graceful SIGTERM->SIGKILL + configurable agent timeout (M-2)

The watchdog used to time.sleep(timeout) then immediately SIGKILL, which cut
claude off mid-write and left half-written artifacts. It now sends SIGTERM,
polls os.kill(pid, 0) for up to agent_kill_grace_seconds, and only SIGKILL if
the process is still alive; ProcessLookupError is tolerated at every step.

Timeout is now configurable via config.py: agent_timeout_seconds (default 1800),
agent_kill_grace_seconds (default 20), and agent_timeout_overrides_json for
per-agent overrides (e.g. {"reviewer": 3600}). AGENT_TIMEOUT is kept as a
backward-compatible alias. The recorded exit_code stays -9 so the ORCH-1
monitor retry/fail logic is unchanged (timeout-kills classify as permanent and
requeue within max_attempts, no retry loop).

src/config.py

@@ -53,6 +53,19 @@ class Settings(BaseSettings):
     breaker_threshold: int = 3
     breaker_pause_seconds: int = 300
 
+    # ORCH-7 (M-2): agent timeout + graceful kill.
+    # agent_timeout_seconds   -> default per-agent wall-clock budget; the watchdog
+    #                            kills the run after this (env ORCH_AGENT_TIMEOUT_SECONDS).
+    # agent_kill_grace_seconds-> pause between SIGTERM and SIGKILL so claude can
+    #                            flush artifacts before the hard kill
+    #                            (env ORCH_AGENT_KILL_GRACE_SECONDS).
+    # agent_timeout_overrides_json -> optional per-agent override JSON object,
+    #                            e.g. {"reviewer": 3600, "architect": 2700}
+    #                            (env ORCH_AGENT_TIMEOUT_OVERRIDES_JSON).
+    agent_timeout_seconds: int = 1800
+    agent_kill_grace_seconds: int = 20
+    agent_timeout_overrides_json: str = ""
+
 
     # Telegram notifications
     telegram_bot_token: str = ""