fix: fail benchmark when cargo nm fails

[0rlych1kk4]

Follow-up to #203.

get_nm() captures the cargo nm subprocess output but does not
currently check its return code. If cargo nm fails, the script can
write an empty nm artifact and continue the benchmark.

This change:

• checks the subprocess return code
• prints the captured stderr
• exits with the same non-zero status

This makes nm-specific failures visible to CI instead of treating them
as valid empty output.

[0rlych1kk4]

CI is passing for the benchmark and other test jobs. The remaining Code Coverage failure is from the Codecov upload returning Token required - not valid tokenless upload, which is unrelated to this change.