Add venv support for custom evals

[krisztianfekete]

This PR adds automatic venv management for evaluators that ship their own dependencies.

• When an evaluator includes a requirements.txt alongside its entrypoint, agentevals now creates an isolated cached venv under ~/.cache/agentevals/venvs/, installs the evaluator SDK and the declared dependencies, and runs the evaluator subprocess using that venv's Python interpreter.
• Each venvs are keyed by evaluator path and invalidated when requirements.txt changes. It prefers uv when available, falling back to venv + pip.

Companion PR: agentevals-dev/evaluators#5 as the first custom evaluator that contains third party dependencies.

Fixes #58

[Copilot]

Pull request overview

Adds optional virtualenv creation/caching for Python custom evaluators that ship a requirements.txt, so evaluator subprocesses can run with isolated dependencies.

Changes:

• Introduce venv management utilities (ensure_venv / ensure_venv_async) that create cached environments and install deps (+ evaluator SDK).
• Extend custom evaluator subprocess execution to optionally run Python evaluators using a venv interpreter.
• Enhance evaluator sources to also fetch/copy requirements.txt alongside evaluator code; update example config to include a deps-heavy evaluator.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
src/agentevals/evaluator/venv.py	New venv creation/install + caching logic for Python evaluators.
src/agentevals/evaluator/sources.py	Fetch/copy requirements.txt alongside evaluator entrypoints.
src/agentevals/custom_evaluators.py	Thread venv interpreter path through runtime command construction and executor factory.
examples/custom_evaluators/eval_config.yaml	Add example evaluator entry intended to exercise dependency installation.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.