Add bertscore eval

[krisztianfekete]

Example config:

evaluators:
  # Built-in metric
  - name: tool_trajectory_avg_score
    type: builtin

  # Custom code evaluators (local scripts)
  - name: bertscore
    type: remote
    source: github
    ref: evaluators/bertscore/bertscore.py
    threshold: 0.7
    timeout: 300
    executor: local
    config:
      expected: "There are two Helm releases installed in the cluster: kagent in namespace kagent (revision 2, deployed, chart kagent-0.7.14) and kagent-crds in namespace kagent (revision 1, deployed, chart kagent-crds-0.7.14)."
      metric: "f1"

Companion PR: agentevals-dev/agentevals#65

Fixes #4