Add regression tests for LTX multi-window fusion and LoRA conversion error paths

[cursor]

What does this PR do?

Consolidates prior regression coverage from cursor/regression-test-coverage-042e (not yet on main) and adds new tests for the highest-risk gaps identified this run.

Risky behavior now covered

• LTX long multi-prompt pipeline — multi-window fusion path: exercises temporal sliding windows with overlap (temporal_overlap > 0, 2+ windows), inter-window tail injection, and linear_overlap_fuse output assembly. Compares multi-window vs single-window latent output to prove the fusion path runs.
• convert_all_state_dict_to_peft error paths: validates the UNet fallback when type inference fails, raises when conversion yields no LoRA keys, and re-raises unexpected errors instead of silently swallowing them.
• Prior branch coverage (042e): CogVideoX schedulers, DDPM Wuerstchen, FlowMatch Euler, Helios, LTX RF, DDIM/DDPM guards, loading_utils, remote_utils, state_dict_utils, and LTX helper unit tests.

Test files added/updated

File	Change
tests/pipelines/ltx/test_ltx_i2v_long_multi_prompt.py	Multi-window inference regression test
tests/others/test_state_dict_utils.py	PEFT conversion error-path tests
tests/others/test_loading_utils.py	New (from 042e)
tests/others/test_remote_utils.py	New (from 042e)
tests/schedulers/test_scheduler_cogvideox.py	New (from 042e)
tests/schedulers/test_scheduler_ddpm_wuerstchen.py	New (from 042e)
tests/schedulers/test_scheduler_flow_match_euler_discrete.py	New (from 042e)
tests/schedulers/test_scheduler_helios.py	New (from 042e)
tests/schedulers/test_scheduler_ltx_euler_ancestral_rf.py	New (from 042e)
tests/schedulers/test_scheduler_ddim.py	Guard test (from 042e)
tests/schedulers/test_scheduler_ddpm.py	Guard test (from 042e)
utils/check_copies.py	Scoped copy check for test-only PRs (from 042e)

Why these tests materially reduce regression risk

• The LTX long pipeline is a complex autoregressive loop; helper tests alone do not prove the multi-window denoising + fusion loop works end-to-end. The new test catches regressions in overlap injection and window stitching that would produce silent shape/continuity bugs.
• convert_all_state_dict_to_peft is a shared LoRA loading utility with fallback logic; incorrect error handling would either crash valid loads or accept invalid state dicts.
• Scheduler and utility tests cover high-blast-radius shared code paths used across many pipelines.

Validation

174 targeted tests pass locally:

python3 -m pytest -n auto --dist=loadfile -s -v \
  tests/others/test_loading_utils.py \
  tests/others/test_remote_utils.py \
  tests/others/test_state_dict_utils.py \
  tests/pipelines/ltx/test_ltx_i2v_long_multi_prompt.py \
  tests/schedulers/test_scheduler_cogvideox.py \
  tests/schedulers/test_scheduler_ddim.py \
  tests/schedulers/test_scheduler_ddpm.py \
  tests/schedulers/test_scheduler_ddpm_wuerstchen.py \
  tests/schedulers/test_scheduler_flow_match_euler_discrete.py \
  tests/schedulers/test_scheduler_helios.py \
  tests/schedulers/test_scheduler_ltx_euler_ancestral_rf.py

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you self-review the diff against .ai/review-rules.md?
• Did you write any new necessary tests?