Empty pruned_init files for three *_task tasks (libero_10_task t2, libero_spatial_task t3/t7) #246

Summary

Three task-perturbation tasks ship pruned_init files with zero initial states. Any loader that picks an init state by index fails on them. As a result, these tasks can't be reset or evaluated. Results on these suites report a denominator of 90 or 80 instead of 100 per suite, which makes them hard to compare.

Reproduction

With the package installed from any of the branches:

from liberopro.liberopro import benchmark

for suite_name, task_id in [("libero_10_task", 2), ("libero_spatial_task", 3), ("libero_spatial_task", 7)]:
    suite = benchmark.get_benchmark_dict()[suite_name]()
    print(suite_name, task_id, len(suite.get_task_init_states(task_id)))
# libero_10_task 2 0
# libero_spatial_task 3 0
# libero_spatial_task 7 0

The BDDL files for these tasks look inconsistent too

These may be why every init state got pruned:

• libero_10_task/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it.bddl:
  the language is "turn on the stove and put the pan on it" and the goal is
  (Turnon flat_stove_1) (On chefmate_8_frypan_1 flat_stove_1_cook_region), but
  the file name refers to the moka pot.
• libero_spatial_task/pick_up_the_black_bowl_on_the_cookie_box_and_place_it_on_the_plate.bddl
  and .../pick_up_the_black_bowl_on_the_stove_and_place_it_on_the_plate.bddl:
  both have the language "Pick the akita black bowl on the top of the cabinet
  and place it on the plate" and the goal (On akita_black_bowl_2 plate_1).
  They differ only in where akita_black_bowl_1 starts (cookies_1 vs.
  flat_stove_1_cook_region), so the two tasks are effectively duplicates.

Questions

1. Can you publish regenerated init states (and
   corrected BDDL if needed)?