feat(evaluator): Add manual e2e script for custom Gym environment validation

[JashG]

Summary

Adds a manual developer e2e test script for exercising custom Gym environments in Kubernetes environment. It validates workstation and cluster prerequisites, prepares either the bundled ASCII Tree example or a caller-provided environment and dataset, uploads the package as a temporary FileSet, configures inference, and submits an Evaluator job that launches the environment inside OpenSandbox. The workflow monitors execution, validates the persisted reward and sandbox lifecycle, records diagnostic evidence, and removes temporary Platform resources when finished.

Changes

• Add Python orchestration for prerequisite checks, environment preparation, evaluation submission, evidence validation, and cleanup.
• Add a bundled ASCII Tree custom environment and wheel-delivered scorer.
• Support custom environment directories, datasets, and resources-server selection.
• Add concise evaluation progress reporting and reusable artifact caching.
• Document cluster setup, Helm configuration, port-forwarding, execution, evidence, and troubleshooting.

Type of Change

• Code change (feature, bug fix, or refactor)
• Code change with documentation updates
• Documentation only
• Contributor tooling or automation
• CI, build, or test infrastructure

Quality Gates

• Tests added or updated for changed behavior
• Existing tests cover changed behavior — justification:
• Tests not applicable — justification: This is a manually invoked, cluster-dependent validation workflow requiring NeMo Platform, OpenSandbox, registry access, and inference credentials. Targeted local checks cover formatting, typing, CLI validation, and environment preparation.
• Documentation updated for user-visible behavior
• Documentation not applicable — justification:

Verification

• Pull request title follows the repository's Conventional Commit format
• Every commit includes an appropriate Signed-off-by: trailer
• uv run pre-commit run -a passes, or any blocked checks are identified below
• Targeted tests pass, or tests are marked not applicable above
• No secrets, API keys, or credentials are included

Targeted validation:

• CLI help and paired custom-input validation passed.
• Environment preparation smoke checks passed for both the bundled and caller-provided input paths.
• I tested the script on a remote cluster using the bundled ascii envrionment and it passed.

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	43616/55571	78.5%	62.2%
Integration Tests	27021/52840	51.1%	22.5%