Add RLVR grader framework, CLI, Lambda deployment, and BFCL tool-calling evaluation

[mccartnick]

feat: Add RLVR grader framework, CLI, Lambda deployment, and BFCL tool-calling evaluation

Issue #, if available: N/A

Description of changes:

Extends llm-eval-kit (v1.0 → v1.1.0) with a grader framework for building and deploying reward functions for Reinforcement Learning with Verifiable Rewards (RLVR) on Amazon Bedrock. All existing SageMaker pre/post processor functionality, tests, and documentation are preserved — this is purely additive.

Grader framework (src/llm_eval_kit/graders/)

• Grader ABC, @grader decorator, and GraderRegistry for defining and registering evaluation functions
• Built-in graders: exact_match, string_similarity (Levenshtein + token F1), tool_call (BFCL-style AST comparison with type coercion and weighted scoring across function name, param names, and param values)

Data models (src/llm_eval_kit/models/)

• Pydantic v2 boundary models: Message, EvalSample, EvalDataset, EvaluateResult, MetricResult
• EvalDataset with JSONL serialization and iteration support

Dataset loaders (src/llm_eval_kit/datasets/)

• load_jsonl — generic JSONL
• load_bfcl — BFCL-specific field mapping
• load_huggingface — pull from HuggingFace Hub with configurable column mapping, data_files selection, and nested list unwrapping for BFCL format

Evaluation pipeline (src/llm_eval_kit/execution/)

• EvalPipeline — runs a grader over a dataset, collects per-sample results, and produces an EvalReport with summary statistics and optional JSONL output

Lambda deployment (src/llm_eval_kit/deploy/)

• deploy_grader — packages a grader + llm_eval_kit + dependencies (pydantic, etc.) into a zip, creates/updates a Lambda function via boto3
• deploy_reward_function — lightweight deployer for standalone zero-dependency reward function files
• YAML-based config with env var overrides and credential chain support
• Prefers uv for dependency installation with pip fallback

CLI (src/llm_eval_kit/cli/)

• llm-eval-kit evaluate — run a grader over a dataset (JSONL or BFCL format)
• llm-eval-kit list-graders — show registered graders
• llm-eval-kit validate — schema-check a dataset file
• llm-eval-kit deploy — deploy a grader as a Lambda reward function

Documentation (docs/)

• graders.md, datasets.md, deploy.md, cli.md
• Updated root README.md with RLVR section and doc links table

Other

• pyproject.toml — added [datasets], [deploy] optional deps, CLI entry point, requires-python >= 3.10
• .gitignore — added for the repo
• uv as the recommended package manager throughout docs and deployment code

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.