Implement on Spot batch VLM predicate evaluation pipeline

[lf-zhao]

Implement a new pipeline for VLM predicate evaluation / VLM classifier:
(Note: The previous PR #284 hasn't been merged and it carries over. That one was mainly for keeping record. This is what actually working on Spot)

Pipeline:

• Initialize VLM predicates in the env, now e.g., On and Inside
• In Spot env reset, Spot builds initial obs and need to see all objects. We provide all VLM predicates and calculate all GroundAtom queries to evaluate for VLM classifiers, and save it as a dictionary.
• In Spot env step, Spot takes one step (skill level) and observes a set of new images, we only update VLM ground atoms of all VLM predicates but evaluate VLMGroundAtoms with the visible objects.
• The evaluation in reset and step are done in batch manner: a batch contains all relevant VLM ground atoms and a collection of Spot camera images (e.g., 6 cameras), and return a list to parse.
• In Spot perceiver (obs -> object-centric state), save VLM predicates and ground atoms to state.
• In abstract (object-centric state -> ground atoms, symbolic state), if it's VLM ground atom, we directly get from state.

Command:
python predicators/main.py --spot_robot_ip 192.168.80.3 --spot_graph_nav_map b45-621 --env lis_spot_block_floor_env --approach spot_wrapper[oracle] --bilevel_plan_without_sim True --seed 0 --perceiver spot_perceiver --spot_vlm_eval_predicate True --vlm_eval_verbose True --num_train_tasks 0 --num_test_tasks 1

Modifications:

• Structs: Add VLM predicate and VLM ground atom class. Partial perception state and Observation class include additional information on visible objects, images, ground atoms from last state, and more.
• Model: Add OpenAI VLM class. Also add some demo VLM predicates based on VLM and their prompts, such as for On, Inside, Blocking.
• VLM predicate evaluation: Add a function to evaluate all visible ground atoms batched in one query.
• Spot VLM object perception: Generate visible ground atoms for visible objects and all VLM predicates.