feat: allow specifying skill_directories per task (#156)

[spboyer]

Summary

Allow skill_directories to be specified per-task in eval YAML files, overriding the eval-level setting. This enables tasks that need different skills from the rest of the eval.

Changes

• internal/models/testcase.go: Add SkillPaths []string field to TestCase struct (yaml:"skill_directories")
• internal/orchestration/runner.go: Update buildExecutionRequest() to prefer task-level SkillPaths over eval-level, falling back when unset
• internal/models/testcase_test.go: Add TestLoadTestCase_SkillDirectories — verifies YAML parsing of the new field
• internal/orchestration/runner_test.go: Add TestBuildExecutionRequest_TaskLevelSkillPaths — verifies override and fallback logic
• schemas/task.schema.json: Add skill_directories array property
• site/src/content/docs/guides/eval-yaml.mdx: Add skill_directories to Task Fields table
• site/src/content/docs/reference/schema.mdx: Add skill_directories section to Task File Format

Behavior

• Task-level skill_directories replaces (not merges with) eval-level paths
• If task has no skill_directories, falls back to eval-level
• Paths resolved relative to the spec directory (same as eval-level)

Example

# eval.yaml
config:
  skill_directories:
    - ./default-skills

tasks:
  - file: special-task.yaml  # can override with its own skill_directories

# special-task.yaml
id: special-task
name: Special task with custom skills
skill_directories:
  - ./custom-skills
inputs:
  prompt: "Do something special"

Closes #156

Working as Linus (Backend Developer)