fix: make grading cross-platform

[paulcam206]

What does this PR do?

Normalizes completion golden-text line endings and makes trajectory shell classification recognize cross-platform read/search commands without treating harmless stderr redirects as mutations.

Why?

Windows CRLF output can fail against equivalent LF golden text, and common PowerShell/CMD inspection commands are misclassified. Read-only grep/find commands that discard or merge stderr are also incorrectly marked as edits.

Changes

• Normalize CRLF and lone CR to LF for inline and file-backed expected stdout comparisons.
• Classify Select-String, recursive/filtered Get-ChildItem, dir /s, Get-Content, and ordinary directory listings precisely.
• Ignore stderr redirects to null or stdout during mutation detection while preserving real file redirects as mutations.
• Add behavior tests for equivalent and different output, Windows command families, and both redirect outcomes.

Tests

• python -m pytest -q passes locally
• python -m pytest -q tests/test_environment.py tests/test_environment_files.py tests/test_trajectory.py
• python -m ruff check clawbench app.py scripts tests passes locally

[clawsweeper]

🦞👀
ClawSweeper picked this up.

Pull request received. I will update this pull request when review starts.