feat: add legacy csv evaluator

[AAgnihotry]

Summary

Add LegacyCSVExactMatchEvaluator for comparing specific CSV columns in agent evaluation outputs.

Features

✅ Single and multiple column comparison - Compare one or more CSV columns (e.g., "Name" or "Name,Age")
✅ Case-insensitive column name matching - "Name", "NAME", "name" all match
✅ Line-by-line evaluation - Evaluate each CSV row separately with header preservation
✅ Automatic job attachment download - Seamlessly handles attachment URIs via base class
✅ Detailed error messages - Clear feedback for missing columns or invalid CSV format

Implementation Details

New Evaluator

• File: src/uipath/eval/evaluators/legacy_csv_exact_match_evaluator.py
• Type: LegacyEvaluatorType.CSVColumnExactMatch (type 10)
• Config: Uses targetSubOutputKey for column names (comma-separated)
• Modes:
  • Normal mode: Compares first data row only
  • Line-by-line mode: Evaluates each row separately with aggregation

Key Implementation Features

1. Base class integration - Uses _get_actual_output() for proper extraction with targetOutputKey
2. CSV parsing - Uses Python's csv.DictReader for reliable parsing
3. Line-by-line support - Preserves headers when splitting rows for evaluation
4. Error handling - Comprehensive validation for missing columns, empty CSV, invalid format

Sample Project

New sample: samples/csv_employee_generator/

• Agent that generates employee CSV data by department
• 6 evaluation test cases covering success, failure, and partial match scenarios
• 3 evaluator configurations (single column, multi-column, line-by-line)
• Uses BaseModel for Input/Output schemas
• Configured to use local repository version via [tool.uv.sources]

Test Coverage

✅ All 1702 tests pass including 22 new CSV evaluator tests

• Single column extraction
• Multiple column extraction
• Line-by-line evaluation
• Job attachment handling
• Error cases (missing columns, empty CSV, invalid format)

Test Results

Engineering - 3 employees:        100%, 100%, 100% ✅
Sales - 2 employees:              100%, 100%, 100% ✅
HR - 2 employees:                 100%, 100%, 100% ✅
Engineering - Wrong name (fail):    0%,   0%,  67% ✅ (correct failure)
Engineering - Partial match:      100%, 100%,  67% ✅ (2/3 rows match)

Files Changed

Core Implementation:

• src/uipath/eval/evaluators/legacy_csv_exact_match_evaluator.py (419 lines)
• tests/cli/evaluators/test_legacy_csv_exact_match_evaluator.py (639 lines)
• Updated evaluator.py discriminator mapping
• Updated models.py with new enum value

Sample Project:

• samples/csv_employee_generator/main.py - CSV generator agent
• samples/csv_employee_generator/evaluations/ - 3 evaluators, 6 test cases
• samples/csv_employee_generator/pyproject.toml - Local package dependencies

Breaking Changes

None - This is a new evaluator type.

Related Issues

Implements CSV column comparison evaluator for agent evaluation framework.

---

🤖 Generated with Claude Code

Co-Authored-By: Claude Sonnet 4.5 noreply@anthropic.com