T-021: Add multi-execution support to exec command

[ali90h]

🚀 T-021: Multi-execution (exec --all/--indexes/--until-success) + JSONL Summary

This PR implements comprehensive multi-execution support for the exec command, enabling users to execute multiple candidate commands from a plan in a single invocation with structured logging and flexible execution control.

✨ New Features

Multi-execution Flags

• --all: Execute all candidate commands in their original order
• --indexes "N,M-P": Execute specific commands by indices/ranges (comma-separated)
• --until-success: Stop after the first command that exits with code 0
• --summary FILE.json: Write final execution summary to JSON file

JSONL Streaming Output

• Run records: One JSONL record per executed command with timing, exit codes, and metadata
• Summary record: Final JSONL record with execution statistics
• Structured logging: Complete audit trail of all executions

Command Selection Logic

• Precedence handling: --indexes > --all > single --index
• Range parsing: Support for ranges like "0,2-4,6" with validation
• Error handling: Clear error messages for invalid configurations

🔧 Implementation Details

Enhanced CLI Interface

• Extended ExecConfig with new multi-execution fields
• Updated argument parsing with comprehensive validation
• Maintained backward compatibility with existing single command execution

JSONL Output Schema

{"type": "run", "index": 0, "cmd": "pytest", "start_ts": "2025-09-13T12:00:00Z", "end_ts": "2025-09-13T12:00:05Z", "exit_code": 1, "duration_ms": 5000}
{"type": "summary", "schema_version": 1, "tool": "autorepro", "runs": 2, "successes": 1, "first_success_index": 1}

Execution Pipeline

• Single execution: Maintains existing behavior for backward compatibility
• Multi-execution: New pipeline with JSONL streaming and summary generation
• Environment handling: Proper setup and cleanup for each command execution

🧪 Testing & Quality

Comprehensive Test Suite

• 19 unit tests covering all new functionality
• Config validation tests: Index parsing, mutual exclusivity, error handling
• Command selection tests: Single/all/indexes selection, precedence rules
• JSONL record tests: Run records, summary records, field validation
• Integration tests: Multi-execution scenarios with proper mocking

Manual Testing

• All acceptance criteria from issue T-021: Multi-execution (exec --all/--indexes/--until-success) + JSONL Summary #111 validated
• Demo plugin with deterministic fail/success/fail pattern
• JSONL output format verification
• Exit code behavior confirmation

📚 Documentation

Updated README.md

• New dedicated "Exec Command (Multi-execution Support)" section
• Comprehensive usage examples for all new flags
• JSONL output format specification
• Multi-execution features explanation
• Complete option reference and exit behavior documentation

CLI Help Text

• Clear descriptions for all new flags
• Usage examples and parameter explanations
• Comprehensive option reference

🔄 Backward Compatibility

• Single command execution: Unchanged behavior and interface
• Existing flags: All existing functionality preserved
• Exit codes: Consistent with existing behavior
• Configuration: No breaking changes to existing config options

🎯 Acceptance Criteria Met

• ✅ --all executes all candidate commands in original order
• ✅ --until-success stops immediately after first successful command
• ✅ --indexes accepts single indices and ranges with proper validation
• ✅ --summary writes JSON file matching final JSONL summary line
• ✅ JSONL output streams per-run and summary records in order
• ✅ Backward compatibility maintained (no JSONL emitted without --jsonl)
• ✅ Invalid indices result in clear error messages and no execution
• ✅ Command precedence properly implemented and tested

🔗 Related Issues

fix #111

📊 Files Changed

• autorepro/cli.py: Core multi-execution implementation
• tests/test_exec_multi.py: Comprehensive test suite (new)
• README.md: Updated documentation
• Cleanup: Removed temporary test files

🚀 Ready for Review

This implementation is production-ready with comprehensive testing, documentation, and maintains full backward compatibility. All linting checks pass and the code follows project standards.