feat(checks): add consensus retries for LLM-based checks

[harsh21234i]

Closes #2372

Summary

• add num_runs and consensus to BaseLLMCheck
• support repeated LLM judge execution with majority, unanimous, and any consensus strategies
• preserve individual run results in details["runs"] when multiple runs are used
• include aggregated consensus metadata in result details
• keep the default single-run behavior unchanged

Testing

• uv run -m pytest -q libs/giskard-checks/tests/builtin/test_base.py libs/giskard-checks/tests/builtin/test_judge.py libs/giskard-checks/tests/builtin/ test_groundedness.py libs/giskard-checks/tests/builtin/test_conformity.py
• uv run -m pytest -q libs/giskard-checks/tests/builtin
• uv run ruff check libs/giskard-checks/src/giskard/checks/judges/base.py libs/giskard-checks/tests/builtin/test_base.py

[gemini-code-assist]

Code Review

This pull request introduces multi-run support for LLM-based checks, allowing users to specify a number of runs and a consensus strategy (majority, unanimous, or any) to determine the final result. The implementation includes logic for aggregating results and selecting representative outputs for the final report. The review feedback suggests a performance optimization to execute the multiple LLM runs in parallel using asyncio.gather instead of the current sequential execution.

[harsh21234i]

hey @kevinmessiaen can you look into this??