feat(train): parallel training framework — FSDP2, HSDP, gradient accumulation, and DCP checkpoints (#4010)

* feat(train): parallel training engine with FSDP2, HSDP, and DCP checkpoints

Replace the FSDP1 training path with a config-owned parallel-training
engine:

- Topology and runtime configs (--parallelism.*, --accelerator.*):
  dp_replicate x dp_shard degrees select single-process, DDP (unchanged
  default), FSDP2, or HSDP; mixed precision, first-class gradient
  accumulation, and FSDP/DDP tuning knobs are mirrored as plain
  dataclasses that build the accelerate objects at runtime, so every
  run is reproducible from its train_config.json alone. Accelerate env
  vars are guarded against configuring the engine behind the config
  system's back.
- Declarative policy surface: policies declare FSDP2 wrap units
  (_fsdp_wrap_modules) and non-forward entry points
  (_fsdp_forward_methods); a shared engine resolves them around
  accelerator.prepare(). Context-parallel fields are reserved and
  validated to 1.
- Checkpoints: selectable --checkpoint_format (safetensors | dcp |
  safetensors_dcp); the sharded optimizer channel is always DCP;
  two-phase resume (step+RNG before prepare, DCP model/optimizer after)
  reshards across GPU-topology changes; lerobot-convert-dcp merges DCP
  shards into a distributable model.safetensors offline.
- Publishing: PreTrainedPolicy.push_model_to_hub is replaced by the
  free publish_trained_model (model + processors + card + train config,
  all-ranks gather with main-rank writes);
  PreTrainedPolicy._save_pretrained gathers state dicts internally,
  removing the state_dict= threading from save_pretrained.
- lerobot_train is restructured around the engine: optimizer built
  before the single prepare() call, deferred weight load on DCP
  resumes, collective save_checkpoint with no call-site rank branches,
  dp-world-size-based sample accounting.

Breaking changes: FSDP checkpoints from lerobot <= 0.6.x are not
resumable (weights stay loadable via from_pretrained; pin
lerobot==0.6.x to finish old runs); the `accelerate launch
--config_file` yaml flow is superseded by the config flags; training
autocast is owned exclusively by --accelerator.mixed_precision
(policy.dtype only casts parameters).

Also fixes: reward-model hub publishing crash (TypeError on extra
kwargs).

Verified by ~200 new CPU tests (config round-trips, checkpoint
round-trips per format, two-phase resume, publisher contracts,
converter equivalence, accelerate canaries), a 5-test 4-GPU suite
(FSDP2 save/resume bit-exactness, HSDP/DDP loss parity,
changed-topology resume, all-ranks save_pretrained, grad-accum
equivalence), and end-to-end ACT (1/4/8 GPUs) + FastWAM 6B
(FSDP2 + HSDP) training runs.

pyproject.toml

@@ -346,6 +346,7 @@ lerobot-record="lerobot.scripts.lerobot_record:main"
 lerobot-replay="lerobot.scripts.lerobot_replay:main"
 lerobot-setup-motors="lerobot.scripts.lerobot_setup_motors:main"
 lerobot-teleoperate="lerobot.scripts.lerobot_teleoperate:main"
+lerobot-convert-dcp="lerobot.scripts.lerobot_convert_dcp:main"
 lerobot-eval="lerobot.scripts.lerobot_eval:main"
 lerobot-train="lerobot.scripts.lerobot_train:main"
 lerobot-train-tokenizer="lerobot.scripts.lerobot_train_tokenizer:main"
@@ -475,6 +476,12 @@ default.extend-ignore-identifiers-re = [
 # TODO: Enable mypy gradually module by module across multiple PRs
 # Uncomment [tool.mypy] first, then uncomment individual module overrides as they get proper type annotations
 
+[tool.pytest.ini_options]
+markers = [
+    "multigpu: distributed tests needing 2-4 GPUs (CI: docker_publish.yml lane)",
+    "multigpu_heavy: 8-GPU sweeps and soak tests; never run in CI",
+]
+
 [tool.mypy]
 python_version = "3.12"
 ignore_missing_imports = true
@@ -521,6 +528,15 @@ disallow_untyped_defs = true
 disallow_incomplete_defs = true
 check_untyped_defs = true
 
+[[tool.mypy.overrides]]
+module = "lerobot.distributed.*"
+ignore_errors = false
+
+# extra strictness for the distributed engine
+disallow_untyped_defs = true
+disallow_incomplete_defs = true
+check_untyped_defs = true
+
 [[tool.mypy.overrides]]
 module = "lerobot.optim.*"
 ignore_errors = false