Replace the pretrained checkpoint functions with CheckpointBundle

[hujc7]

Summary

Stacked on #7485 — [Fix] Publish and fetch the checkpoints a task's components declare, which is the
minimal fix. This PR is the refactor that PR deliberately kept out of scope.

• One CheckpointBundle replaces 13 free functions that each took the same
  (workflow, task_name, physics_backend, render_backend) tuple; the publish script unpacked it at 20 call sites.
• One Workflow per RL library replaces four parallel tables and the 13 resolve_checkpoint_selector blocks
  that each spelled that library's file patterns. Two copies had drifted: the benchmark sb3 pattern excluded
  model.zip, so its own preference never matched; skrl preferred best_agent.pt only when publishing.
• Checkpoint.local_path replaces the cache-dir-as-log-dir hand-off, which put rl_games and skrl companions
  one directory off.

Description

1. Change

• isaaclab.utils: Checkpoint.local_path, set by the fetch and returned first by resolve(); find_cfgs
  and latest_file become public utilities.
• isaaclab_rl.utils.pretrained_checkpoint: Workflow (policy glob, preferred file, extension, experiment
  key; selector_args() derives the selector patterns) in WORKFLOWS; CheckpointBundle owns identity,
  companions and the published, cached and collected paths, and fetch() records each companion's
  local_path. get_published_pretrained_checkpoint(…, env_cfg=) is unchanged, so its ten call sites are
  untouched.
• Entrypoints (4 play, 4 train, 4 benchmark, leapp sb3 export): **WORKFLOWS[lib].selector_args(…).
• train_and_publish_checkpoints.py: CheckpointJob owns training-run state and the train/play commands;
  from_task resolves the preset-selected config so the backend names and declared checkpoints describe
  what the presets train; the trained policy is selected through the same manifest-based lookup as
  --checkpoint best.

2. Migration

Full retired→new table in source/isaaclab_rl/changelog.d/checkpoint-bundle.major.rst. Behaviour
changes: skrl --checkpoint best prefers best_agent.pt; the publish script only collects runs carrying
a run.json manifest; --publish_root keeps the legacy per-task sub-directory.

3. Verification

200 pass across test_checkpoints.py, test_pretrained_checkpoint.py, test_train_and_publish_checkpoints.py,
test_shadow_hand_camera_presets.py, test_pretrained_checkpoint_lookup.py, test_standalone_scripts.py,
test_entrypoints.py, test_checkpoint_manifest.py and benchmark test_api.py. New tests pin each library's
selector patterns against the files it writes, the local_path write-back, manifest-required selection and
the preset-config plumbing of from_task. pre-commit clean.

Type of change

• Breaking change (existing functionality will not work without user modification)

Release backport

• Backport this pull request to the active release branch after it merges into develop

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• Documentation needs no change: the module has no API page; the changelog fragment carries the migration table
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[hujc7]

run-ci

[hujc7]

run-ci

[hujc7]

run-ci