Add local ART-E email agent example

[poofeth]

Summary

• add a local examples/art-e Python package for an ART-E style email research agent
• include deterministic inbox fixtures, search/read tools, LangGraph rollout, ART training entrypoint, and scoring helpers
• add provider-free rollout/train smoke tests that monkeypatch LangGraph model creation and verify ART-E env handling
• make Weave tracing optional for this example so local smoke tests can run even if Weave's transitive gql import is incompatible in a fresh environment
• document offline checks/training and link the local example from the root README

Scope

This is a local deterministic ART-E-style example. The tests mock model calls and require no Gmail account, paid API, or private credentials. Real training uses art_e.train with a configured inference endpoint.

Bounty

• Algora: https://algora.io/PrimeIntellect-ai/bounties/cxptwFoEv2up9Qjm
• This PR is submitted as the Easy lane claim for the ART-E / ART custom Algora bounty.
• This targets the easy lane described as an ART-E agent task.

Validation

• uv run --project examples/art-e pytest examples/art-e/tests -q - 6 passed
• (cd examples/art-e && uv run python main.py)
• uv run --project examples/art-e ruff check .
• uv run --project examples/art-e ruff format --check .
• python -m compileall examples/art-e/art_e examples/art-e/main.py examples/art-e/tests
• git diff --check

[poofeth]

Added provider-free smoke coverage for the ART-E entrypoints: the new test monkeypatches LangGraph chat/model creation, drives the final-answer tool, verifies reward/scoring, and checks build_model reads the ART-E env vars. Validation rerun: uv run --project examples/art-e pytest examples/art-e/tests -q (6 passed), (cd examples/art-e && uv run python main.py), ruff check/format-check, compileall, and git diff --check.

[poofeth]

Removed the generated per-example uv.lock from the PR to keep the example diff focused; this cuts 6,619 generated lines without changing the source package.\n\nReran validation after removal:\n\ntext\n$ uv run --project examples/art-e pytest examples/art-e/tests -q\n...... [100%]\n6 passed, 7 warnings in 3.01s\n\n$ (cd examples/art-e && uv run python main.py)\nScenario: Where and when is the team offsite?\nRetrieved: Team offsite logistics\nReward: 0.93\n\n$ uv run --project examples/art-e ruff check .\nAll checks passed!\n\n$ uv run --project examples/art-e ruff format --check .\n311 files already formatted\n\n\nAlso updated the PR body with a scope note: tests are deterministic/provider-free; real training requires a configured inference endpoint.

[poofeth]

/claim https://algora.io/PrimeIntellect-ai/bounties/cxptwFoEv2up9Qjm