[Kewe63]

Description

• Preserve every requested load-test broadcast attempt while tracking the first nonzero forge status.
• Exit explicitly with the accumulated status after the loop.
• Log the failed iteration and status for operator visibility.
• Add a dependency-free Bash regression with a PATH-local fake forge.

Motivation and Context

The wrapper previously exposed only the final loop command's status. An earlier failed broadcast followed by a successful final broadcast therefore made the whole load test exit 0.

The fix retains the wrapper's load-generator behavior by continuing all requested iterations, but it can no longer report success if any broadcast failed.

Fixes #251.

The separate zero-run validation issue is deliberately left out of this PR.

How Has This Been Tested?

• Pre-fix regression — failed as expected: intermediate fake broadcast returned 42, all 3 requested calls ran, but the wrapper returned 0.
• bash script/test/7683/loadtesting/loadtest.failure-status.test.sh after the fix — passed:
  • all-success: status 0, 3 calls;
  • intermediate failure: status 42, 3 calls.
  • multiple failures: preserved the first status (41), 3 calls.
  • failure log includes the propagated status.
• bash -n on the production wrapper and regression script — passed.
• git diff --cached --check — passed.
• Independent code review — passed with no blocking correctness, shell, security, or test findings; signal-derived status 143 propagation was also verified.
• The regression uses a PATH-local fake forge, placeholder environment values, and an isolated temporary directory; it performs no RPC or chain operation.

Types of changes (remove all unchecked types)

• Bug fix (non-breaking change which fixes an issue)
• Test coverage

Checklist:

• I assessed documentation impact; no documentation update is required.
• I assessed deployment-script impact; no deployment script is changed.
• I kept RUN_COUNT validation unchanged for the separate zero-run issue.