Split BATS test-task contract from runner implementation lint

[c0da-ricon]

Problem

lint:bats-test-task currently conflates the public mise task contract with the runner implementation.

During review of KnickKnackLabs/template#8, a readability fix moved the nontrivial BATS workflow from .mise/tasks/test into a repo-local libexec/test, leaving the public task as a thin Usage/argument adapter. The real task path and BATS behavior remain covered by integration tests, but codebase 0.3.0 fails with:

no 'bats' invocation found

The rule currently checks several distinct concerns together:

• #USAGE argument and example metadata;
• literal direct BATS invocation inside .mise/tasks/test;
• invocation count;
• implementation anti-patterns such as an outer if $# wrapper.

That makes a direct invocation an architectural requirement even when a checked repo deliberately uses a thin task adapter and explicit internal runner.

Direction

Separate the contracts, or otherwise make delegation first-class.

A likely split is:

• bats-test-task: validate the public test-task interface and Usage metadata;
• bats-test-runner: validate the direct or explicitly delegated runner implementation.

For delegated runners, keep the boundary narrow and safe:

• delegation must use a literal repo-local path;
• the target must exist, be executable, remain inside the repo, and not be a symlink;
• exactly one BATS execution should exist across the adapter and runner;
• existing runner anti-pattern checks should still apply;
• dynamic or outside-repo delegation should not pass accidentally.

Current consumer boundary

Template will temporarily remove bats-test-task from its configured lint set rather than add an ignore directive or retain dense task implementation solely to satisfy the current rule.

This issue does not require Template#8 to wait for a codebase release. It tracks restoring an accurate generic lint contract afterward.

Acceptance

• Direct canonical BATS tasks continue to pass.
• Thin task adapters with safe explicit delegated runners can be checked.
• Missing Usage metadata still fails at the task boundary.
• Missing, dynamic, symlinked, outside-repo, or multiply-invoked runners fail clearly.
• Existing fixtures and diagnostics are reorganized around the separated concerns.
• Fold's BATS/code-structure guidance is updated if the public rule names or contract change.