checks: add retrieval quality checks

[harsh21234i]

Closes #2445

Summary

• Add built-in strict retrieval checks: RecallAtK, PrecisionAtK, HitRateAtK, MRR, NDCGAtK, and InfAP
• Support configurable threshold, JSONPath keys for relevant/retrieved IDs, and k where applicable
• Export the new checks from giskard.checks
• Add unit coverage for empty inputs, perfect retrieval, partial overlap, ranking sensitivity, sparse-label InfAP,
  duplicate retrieved IDs, missing keys, and registry validation

Scope

This PR implements the strict exact-ID matching strategy first. Cosine similarity, LLM-judged relevance, and
documentation updates can be added in follow-up PRs.

Testing

• uv run -m pytest -q libs/giskard-checks/tests/builtin/test_retrieval.py
• uv run -m pytest -q libs/giskard-checks/tests/builtin
• uv run ruff check ...
• uv run basedpyright ...

[gemini-code-assist]

Code Review

This pull request introduces a comprehensive set of retrieval quality metrics, including Recall@K, Precision@K, HitRate@K, MRR, NDCG@K, and InfAP, along with corresponding unit tests. Feedback suggests refining the _as_sequence helper to handle None values correctly, adjusting the Precision@K calculation to use a standard denominator, and renaming the InfAP metric to AveragePrecision for better alignment with information retrieval terminology.

[gemini-code-assist]

The metric implemented here is standard Average Precision (AP). In IR literature, Inferred Average Precision (InfAP) refers to a specific estimator designed for incomplete relevance judgments (where some documents are unjudged). Since this implementation assumes strict exact-ID matching against a provided set (complete judgment), it should be renamed to AveragePrecision to avoid confusion with the specialized InfAP metric.