fix(exec): per-step seconds in the run summary and BUILD-LOG, gate seconds kept on a failed gate (409a)

[MendixMau]

Direct lane

What changed and why (one paragraph):
exec.sh now prints and logs per-step seconds: steps: guards Ns check Ns snapshot Ns baseline Ns exec Ns gate Ns lint Ns score Ns in the run summary and at the end of the BUILD-LOG detail cell. Before this, a row carried only the run total, so nobody could say where a 28 to 41 s exec went. The gate mark is taken straight after the mxbuild run, before any pass or fail branch. In the first version (in #409) gate was marked only inside log_applied, so every rolled-back or failed-gate row lost its gate seconds; in the lab those are 47 of 238 rows, the costliest 20%. Split out of draft #409 (first of three) so that both arms of the test-strategy A/B log the same steps: cell.

Impact

Row	Answer
Process	none. No new step, gate, blocking path or artifact. One extra cell in the BUILD-LOG detail.
Dependencies	Shared file: project-bin/exec.sh (marks only; no behaviour change). Parsers that read the BUILD-LOG detail: grepped bin/ and project-bin/, none reads the detail by position, and the new cell is last. No new artifact reads. Independent of the --subjects PR (b); both prepend a CHANGELOG line, so the second to merge needs a trivial CHANGELOG rebase.
Context cost	none: no routed or always-loaded file touched
Behaviour	n/a: scripts only
Runtime	8 to 9 date +%s calls per exec. Measured: one call takes 22.8 ms under bash 3.2, so about 0.2 s per exec, 0.5 to 0.7% of a 28 to 41 s exec.

Field evidence: the per-step split and the failed-gate loss were measured on a five-app lab project (Mendix 11.15, mxcli v0.26.0, 238 BUILD-LOG rows).

• No client data anywhere in the diff
• Size cap: 2 files, 111 lines (17 code, 94 test), plus CHANGELOG
• Test tier: T1, scoped fixture. tests/wave2/test-exec-step-times.sh 7/0 on this branch, 4 pass and 3 fail on master's exec.sh (summary steps line, BUILD-LOG steps cell, failed-gate row). It covers a passing and a gate-failing --patch run. Existing: test-exec-change-class.sh 21/0, test-exec-lock-watch.sh 84/0, test-manifest-clean-checkout.sh 1/0. Only macOS was run; Git Bash was not.
• CHANGELOG.md line, credited
• Categorisation: kind fix · lane Direct · size S · regression risk low · toolkit change, not an upstream mxcli bug

🤖 Generated with Claude Code