Makefile: run every benchmark platform and report failures to Buildkite

benchmark-platforms runs the platforms in a set -e loop and does not capture each platform's exit status, so the first platform that fails aborts the loop and the remaining platforms never run. Today kvm fails on every run (BenchmarkRestore* is broken on Docker 28, moby/moby#50750), so systrap never runs and systrap data is never populated.

The failure also never surfaced in Buildkite. The benchmark steps run make -i, which ignores the error and exits 0, so Buildkite marked the step as passed and the step's soft_fail: true was never used.

Changes:

• benchmark-platforms, benchmark-platforms-nodocker: capture each platform's exit status (|| exit_code=$?), run all platforms and runc, then exit $exit_code.
• run_platform_benchmark: on failure, still upload the parsed results and remove the log file, and return the benchmark's exit status (same structure as run_benchmark).
• .buildkite/pipeline.yaml: remove -i from the benchmark steps. The Makefile now continues past failures.

Effect: every platform runs and uploads its passing benchmarks even if an earlier platform failed. A step with a failing benchmark exits non-zero, which soft_fail: true turns into a yellow soft-failed step; the build still passes. Until BenchmarkRestore* is fixed, the Startup step will be yellow on every run.
FUTURE_COPYBARA_INTEGRATE_REVIEW=#16078 from riverafernando:fix-benchmark-platforms 79ed273