fix(exec): keep the gate's full mxbuild output in a per-run log

[MendixMau]

Problem

exec.sh printed head -20 of mxbuild's output and deleted the rest with its temp file, so the evidence of a failed gate (the detail of a CE1613 among it) was lost. In a field migration project (12 modules), every one of the 17 "gate could not run" rows and at least 1 real CE failure had no output left to read, which cost about 20 minutes of blocked reruns.

Refs #442, #448, #171, #285, #424

Change

• project-bin/exec.sh: when the gate does not end clean (non-zero exit or a non-empty errors file) the full mxbuild output is written to .mpr-snapshots/mxbuild-logs/<timestamp>-<pid>.log. That dir is git-ignored by snapshot-mpr.sh, and its name does not match the [0-9]* pattern restore_snapshot uses, so a rollback never touches it. The newest 30 logs are kept.
• The screen keeps the top lines and adds full mxbuild output: <path>; the BUILD-LOG row detail ends with mxbuild log: <path> (the detail column is free text, so the row format is unchanged).
• CHANGELOG entry.

Test

New tests/wave2/test-exec-gate-log.sh (stub mxcli, mxbuild, java; no real mxbuild): A) a failing gate with 41 lines of output leaves a log holding all of it, the screen shows no tail and names the path, the row names the path, and the rollback still applies; B) a clean run writes no log and no note; C) 35 old logs plus one new are pruned to 30. 11 of 11 pass. Existing test-exec-crash-rollback.sh (13) and test-exec-lock-watch.sh (84) pass. check-pr-discipline.sh and check-portability.sh (changed files) are clean.

Risk

• A failed gate writes one log file (size of mxbuild's output); clean gates write nothing.
• Touches the same BUILD-LOG detail line as the watcher PR (fix/gate-watcher-race); a one-line rebase conflict is possible.

🤖 Generated with Claude Code