fix(jobs): address claude review findings on remote training

Resolve the claude[bot] review on #3856:

- Reject reward-model training under --job.target with a clear error instead
  of crashing on a None policy inside build_remote_config_file.
- Support --policy.path remote runs: validate() no longer requires repo_id for
  remote runs (it is auto-generated in submit_to_hf), and repo_id/push_to_hub
  are now set after validate() resolves the policy.
- Narrow the bare `except Exception` in _tail_logs/_poll_until_done to
  (OSError, httpx.HTTPError) so programming errors surface instead of being
  silently retried or counted as job failures.
- Install the SIGINT detach handler only on the main thread.
- Generate model repo timestamps in UTC.

tests/jobs/test_hf.py

@@ -18,6 +18,7 @@ import threading
 from types import SimpleNamespace
 
 import draccus
+import httpx
 import pytest
 
 from lerobot.configs.train import TrainPipelineConfig
@@ -58,9 +59,9 @@ def test_poll_until_done_exits_when_done_already_set(monkeypatch):
     assert _poll_until_done("j", done, poll_interval=0.01) is None
 
 
-def test_poll_until_done_gives_up_after_repeated_failures(monkeypatch):
+def test_poll_until_done_gives_up_after_repeated_network_failures(monkeypatch):
     monkeypatch.setattr(
-        "lerobot.jobs.hf.inspect_job", lambda job_id: (_ for _ in ()).throw(RuntimeError("boom"))
+        "lerobot.jobs.hf.inspect_job", lambda job_id: (_ for _ in ()).throw(httpx.ConnectError("boom"))
     )
     done = threading.Event()
     result = _poll_until_done("j", done, poll_interval=0.001, max_failures=3)
@@ -68,6 +69,14 @@ def test_poll_until_done_gives_up_after_repeated_failures(monkeypatch):
     assert done.is_set()
 
 
+def test_poll_until_done_propagates_programming_errors(monkeypatch):
+    """A bug (e.g. TypeError) must surface, not be silently retried as a transient failure."""
+    monkeypatch.setattr("lerobot.jobs.hf.inspect_job", lambda job_id: (_ for _ in ()).throw(TypeError("bug")))
+    done = threading.Event()
+    with pytest.raises(TypeError):
+        _poll_until_done("j", done, poll_interval=0.001, max_failures=3)
+
+
 def test_resolve_wandb_key_from_env(monkeypatch):
     monkeypatch.setenv("WANDB_API_KEY", "abc123")
     assert resolve_wandb_api_key() == "abc123"
@@ -121,6 +130,23 @@ def _minimal_cfg():
     )
 
 
+def test_validate_skips_repo_id_check_for_remote():
+    """Remote runs auto-assign repo_id in submit_to_hf, so validate() must not demand it up front."""
+    cfg = _minimal_cfg()  # remote target, push_to_hub default True, no explicit repo_id
+    assert cfg.policy.repo_id is None
+    cfg.validate()  # must not raise
+
+
+def test_validate_requires_repo_id_for_local_push():
+    """Local runs that push to the Hub still need an explicit repo_id."""
+    cfg = draccus.parse(
+        TrainPipelineConfig,
+        args=["--dataset.repo_id", "u/d", "--policy.type", "act"],
+    )
+    with pytest.raises(ValueError, match="repo_id"):
+        cfg.validate()
+
+
 def test_build_remote_config_applies_overrides(tmp_path):
     cfg = _minimal_cfg()
     dest = tmp_path / "train_config.json"
@@ -192,7 +218,7 @@ def test_submit_requires_login(monkeypatch):
 
 
 def test_submit_passes_validation_and_submits(monkeypatch):
-    """Regression: repo_id must be set BEFORE cfg.validate() or validation raises."""
+    """A type-based policy with no explicit repo_id is auto-assigned one and submitted."""
     from unittest.mock import MagicMock
 
     # Patch get_token
@@ -261,6 +287,27 @@ def test_submit_passes_validation_and_submits(monkeypatch):
     assert call["labels"].get("lerobot") == "true"
 
 
+def test_submit_rejects_reward_model_training(monkeypatch):
+    """Remote training only supports policies; reward-model runs fail fast with a clear error."""
+    monkeypatch.setattr("lerobot.jobs.hf.get_token", lambda: "tok")
+
+    class FakeHfApi:
+        def __init__(self, token=None):
+            pass
+
+        def whoami(self, token=None):
+            return {"name": "alice"}
+
+    monkeypatch.setattr("lerobot.jobs.hf.HfApi", FakeHfApi)
+
+    cfg = _minimal_cfg()
+    cfg.reward_model = SimpleNamespace(type="reward")  # marks this as reward-model training
+    monkeypatch.setattr(cfg, "validate", lambda: None)  # skip pretrained-path resolution
+
+    with pytest.raises(ValueError, match="reward model"):
+        submit_to_hf(cfg)
+
+
 @pytest.mark.timeout(15)
 def test_submit_returns_when_job_completes(monkeypatch):
     """Non-detach path must RETURN (not hang) once the job reaches a terminal stage."""