fix(pi07_paligemma): address review on high-level planner

[shuheng-liu]

What this does

Addresses review feedback on PR #167 for the π07 PaliGemma high-level planner:

• Register pi07_paligemma_high_level_planner and pi07_paligemma_low_level_planner in src/opentau/policies/factory.py.
• Remove the init_strategy config from PI07HighLevelPlannerConfig (no action expert here, it was a debug-only utility); hardcode load_pretrained_paligemma=False and drop the _init_model / _init_weights indirection.
• prepare_metadata: switch zip to strict=True and assemble the metadata via " ".join(segments) to avoid trailing whitespace when "mistake" is absent.
• Rename prepare_memory → prepare_next_memory and read the target memory from batch["next_memory"] (since batch["memory"] is conventionally the current memory). Update the call site, docstring, and integration test fixture accordingly.
• Comment fix: clarify the "if '' is found" comment in the renamed memory method.

How it was tested

• pre-commit run --files <changed files> — all hooks pass (ruff, ruff-format, pyupgrade, typos, bandit, license header, etc.).
• Integration tests in tests/policies/test_pi07_paligemma_high_level_planner.py are GPU-only (@pytest.mark.gpu, @pytest.mark.slow) and were not re-run in this environment; the test fixture was updated to use the new next_memory key.

How to checkout & try? (for the reviewer)

pytest -sx tests/policies/test_pi07_paligemma_high_level_planner.py -m "gpu"

Checklist

• I have added Google-style docstrings to important functions and ensured function parameters are typed.
• My PR includes policy-related changes.
  • If the above is checked: I have run the GPU pytests (pytest -m "gpu") and regression tests.

Note: Before submitting this PR, please read the contributor guideline.

---

Generated by Claude Code