Add regression tests for schedulers, LoRA utils, and remote decode

[cursor]

What does this PR do?

Adds fast unit tests covering high-blast-radius production paths that previously had no dedicated scheduler or utility coverage.

Risky behavior now covered

• FlowMatchEulerDiscreteScheduler — shared by SD3, Flux, Wan, and other flow-matching pipelines (timestep setup, dynamic shifting guard, step contract, denoising loop)
• LTXEulerAncestralRFScheduler — LTX long-form video pipelines (custom sigma schedules, eta=0 determinism, step API)
• HeliosScheduler — multi-stage set_timesteps(stage_index=...), Euler step, UniPC corrector path on second step
• DDIMScheduler / DDPMScheduler — set_timesteps guard when num_inference_steps > num_train_timesteps (prevents inverted comparison regressions)
• state_dict_utils — LoRA key remapping (PEFT ↔ Diffusers ↔ Kohya prefixes and alpha tensors)
• remote_utils — decode/encode validation, image-type detection, postprocessing

Test files added/updated

File	Tests
tests/schedulers/test_scheduler_flow_match_euler_discrete.py	new — 9 tests
tests/schedulers/test_scheduler_ltx_euler_ancestral_rf.py	new — 10 tests
tests/schedulers/test_scheduler_helios.py	new — 8 tests
tests/others/test_state_dict_utils.py	new — 11 tests
tests/others/test_remote_utils.py	new — 9 tests
tests/schedulers/test_scheduler_ddim.py	+2 guard tests
tests/schedulers/test_scheduler_ddpm.py	+2 guard tests

Why these tests materially reduce regression risk

• Schedulers with non-standard APIs cannot use SchedulerCommonTest; without dedicated contract tests, step-signature or timestep-schedule bugs slip through pipeline-only coverage.
• LoRA conversion utilities are on every load/save path — incorrect key remapping silently produces wrong weights.
• set_timesteps guard tests lock in correct comparison direction; an inverted check would reject all valid schedules.
• Remote decode helpers gate server-side tensor round-trips; validation errors here surface only at integration time.

Validation

All 49 targeted tests pass locally:

python3 -m pytest -n auto --dist=loadfile -s -v \
  tests/schedulers/test_scheduler_flow_match_euler_discrete.py \
  tests/schedulers/test_scheduler_helios.py \
  tests/schedulers/test_scheduler_ltx_euler_ancestral_rf.py \
  tests/others/test_state_dict_utils.py \
  tests/others/test_remote_utils.py \
  tests/schedulers/test_scheduler_ddim.py::DDIMSchedulerTest::test_set_timesteps_num_inference_steps_exceeds_train_timesteps_raises \
  tests/schedulers/test_scheduler_ddim.py::DDIMSchedulerTest::test_set_timesteps_num_inference_steps_at_limit_succeeds \
  tests/schedulers/test_scheduler_ddpm.py::DDPMSchedulerTest::test_set_timesteps_num_inference_steps_exceeds_train_timesteps_raises \
  tests/schedulers/test_scheduler_ddpm.py::DDPMSchedulerTest::test_set_timesteps_num_inference_steps_at_limit_succeeds

Before submitting

• Did you write any new necessary tests?