Tasks: four `test_instance.py` tests read the machine they run on

Phase 4 of 4. Each task names the acceptance criterion it satisfies and the testing-plan row that proves it. Ordered as the ticket asks: the bug first, the parity second, so the parity fix is not what hides the next divergence.

[x] A0 — reproduce, and pin the negative control. Run the four tests in all four rows of the ticket's table on main; confirm 4 failed / 59 passed in three of them. AC: AC1 · Test: T5.
[x] A1 — pin the lookup where the defect is. cli/tests/test_instance.py gains the autouse _no_ambient_cli_config fixture, setting $THE_LOOP_CLI_CONFIG to an uncreated path under the test's own tmp_path, so default_cli_config_path() and _cli_config_export() still run for real and find nothing. AC: AC1 · Test: T1, T11.
[x] A2 — establish the scope by trying the wider one. The same fixture in cli/tests/conftest.py, suite-wide, then run T2. It fails 14 tests whose state root is anchored on the config path; revert to A1 and record why in bugfix.md § Defect 1 and evidence/verification.md. AC: AC1 · Test: T2.
[x] B1 — make the Makefile's parity claim true. test: becomes cd cli && uv run python -m pytest -q, byte-for-byte the pre-commit pytest hook's entry; a # CI parity: comment says why the cd is part of the command; the header RULE is narrowed from "all scripts run from the project root" to invocation, naming the exception. Deps: A1 (the ticket's order) · AC: AC2, AC3 · Test: T4.
[x] C1 — docs. docs/capabilities/testing-and-contracts.md gains the standing rule (a test pins or controls what it resolves; a cwd-relative lookup is not ambient in itself and is not pinned wholesale; a Makefile parity claim includes the working directory) and a history row. Deps: A1, B1 · Test: T4 (markdownlint via make lint).
[x] D1 — evidence. verification.md (all matrix rows with raw output, including the rejected suite-wide variant), self-review.md, security-review.md, documentation.md, reviewer-briefing.md. Deps: A1–C1.