[ENG-4231] Add CLI support for verifiers v1 env configs

[eexwhyzee]

Note

Medium Risk
Changes how training/eval environment payloads are validated and serialized before run creation; mistakes could reject valid configs or send wrong API shapes, but scope is CLI config parsing with new tests covering v1 paths.

Overview
Adds verifiers v1 environment blocks to prime train TOML parsing and API payloads while keeping legacy [[env]].id configs working.

Training and eval [[env]] / [[eval.env]] entries can use taskset + harness (and related v1 fields like group_size, ratio, pool/timeouts, token caps) without a top-level id. Config validation requires either legacy id or taskset.id, validates Hub-style owner/name refs, and maps eval rollouts_per_example → group_size for v1 envs.

The init template documents the v1 shape. Pre-run Hub action checks run only for slash-shaped refs (legacy id, taskset.id, harness.id), not bare runtime ids like alphabet-sort-v1. Run summaries and lists use clearer display names for v1 configs.

^{Reviewed by Cursor Bugbot for commit 7dc15bb. Bugbot is set up for automated code reviews on this repo. Configure here.}

[eexwhyzee]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. 👍

Reviewed commit: 7dc15bb7da

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".