Improve classify workflow

[amrit110]

This pull request simplifies and streamlines the workflow for classifying pull request failures and analyzing logs in the fix-remote-pr.yml GitHub Actions workflow. Key changes include removing custom shell logic for fetching and parsing failure logs, delegating more responsibility to the aieng-bot classify CLI, and improving JSON parsing robustness in the Python classifier. The workflow is now less error-prone and easier to maintain.

Workflow simplification and delegation:

• Removed all custom shell logic for fetching, extracting, and copying failure logs in .github/workflows/fix-remote-pr.yml; the workflow now relies on the aieng-bot classify CLI to fetch PR information and logs directly, reducing complexity and maintenance overhead. [1] [2] [3] [4]
• Updated the classification step to call aieng-bot classify with just the repo and PR number, and to parse output JSON for results, removing manual environment variable passing and output parsing.
• Removed all references to .failure-logs.txt from file copying, git exclude lists, and cleanup steps, since logs are no longer handled by shell code. [1] [2] [3] [4] [5]

Python classifier improvements:

• Added a new strategy in src/aieng_bot/classifier/classifier.py to attempt to fix and parse incomplete JSON responses by adding missing closing braces, making the classifier more robust to partial model outputs.
• Improved logging and warning when Claude's response is truncated due to max_tokens, and refactored loop variable naming for clarity. [1] [2]

Dependency updates:

• Added python-dotenv as a required dependency in pyproject.toml for improved environment variable management.