[BUG] --trials is silently ignored without --matrix

[Benjamin-eecs]

Required prerequisites

• I have searched the Issue Tracker and this has not been reported before.
• I have verified the problem against the latest main (ff10b7d).

What version of BenchFlow are you using?

0.7.6.dev2102 (PyPI); code paths verified identical on main at ff10b7d.

System information

Host: macOS 15 (arm64), Python 3.12.13. Sandbox: Daytona and Docker (behavior is sandbox-independent).

Problem description

bench eval run --trials N is only consumed by the --matrix expansion. A plain run accepts the flag, silently ignores it, and executes one trial per task. The help text says "Number of trials for --matrix" (src/benchflow/cli/main.py:615), but nothing stops or warns the non-matrix caller, so a batch believed to be N trials produces N=1 data.

We ran a full 16-task x 3-model batch with --trials 3 and got exactly one trial per (task, model) cell before noticing.

Reproducible example code

bench eval run --tasks-dir <dir> --agent oracle --sandbox docker --trials 3

Traceback

No error; exit is clean and the jobs dir contains one rollout per task.

Expected behavior

--trials > 1 without --matrix fails fast with a clean EvalPlanError, the same pattern as the neighboring flag-dependency guards (--registry requires --dataset, --matrix currently requires --tasks-dir).

Additional context

Fix proposed in #1064.