fix(ci): stop green-washing coverage-floor failures in suite_runner

[docxology]

The defect

infrastructure/reporting/suite_runner.py forced exit_code = 0 whenever check_test_failures returned should_halt=False. But that helper returns should_halt=False whenever failed_count == 0 — and a --cov-fail-under failure is exactly every test passing while pytest exits 1. So a project dropping below its coverage floor reported PASS. (Found by Forge during the template_newspaper verification; empirically reproduced by revert-and-rerun.)

The fix

Only suppress the non-zero exit for tolerated test failures (failed_count > 0 within the configured max). A non-zero exit with zero test failures — coverage-floor failure or an internal pytest error — keeps its exit code.

Two regression tests pin both paths:

• test_coverage_floor_failure_is_not_green_washed (exit 1 + 0 failed → stays 1)
• test_tolerated_test_failures_are_still_suppressed (exit 1 + 2 failed, max 5 → suppressed to 0)

Coupled fix: a previously-vacuous test

test_run_infrastructure_tests_pipeline_smoke_real_subprocess only passed because of the green-wash: its synthetic repo couldn't import infrastructure, so the smoke subprocess exited 2 (collection error) and was masked to 0 — it never actually ran anything. It now seeds trivial passing tests at the smoke paths, genuinely exercising the runner's path-resolution + subprocess + exit-code propagation. (Symlinking the real smoke tests cascades into a conftest/helper dependency chain a synthetic repo can't satisfy.)

Verification (run, not asserted)

• Targeted: 32 infra tests pass (incl. both regressions + the repaired smoke test); ruff/format/mypy clean.
• Full infra suite: 7178 passed, 0 failed — confirms no other test was relying on the masking.