chore: rename dataset → eval_set and derive from name field

[christso]

Objective

Rename the dataset field to eval_set throughout the codebase and change its derivation to use the YAML name field instead of the old dataset field or filename fallback.

Motivation

• Align with agentv-bench which already uses eval_set terminology
• dataset is generic and overloaded; eval_set is more precise
• Many eval files share the same filename (dataset.eval.yaml), making filename-derived dataset values meaningless — the name field is a better source

Design

• Rename: dataset → eval_set in types (EvalTest, EvaluationResult), Zod schema, orchestrator, CLI commands, OTel exporter, tests, and JSONL baselines
• Derivation change: eval_set reads from suite.name (already in schema) instead of suite.dataset, falling back to filename
• Clean break: Remove dataset from RawTestSuite — no backward compat shim
• CLI: --group-by dataset becomes --group-by eval-set
• Examples: Add name: field to example YAMLs that only had description:

Acceptance Signals

• All tests pass with renamed field
• eval_set in results JSONL output reflects name from YAML (not filename)
• CLI --group-by eval-set works
• No remaining references to dataset as a field name in core/CLI code

Non-Goals

• Renaming filenames (e.g., dataset.eval.yaml stays — it's a file naming convention)
• Adding name: to every example YAML (just representative ones)

Related

• PR chore: rename dataset to eval_set, derive from YAML name field #678