benchmark.sh: apply PR #202 review-thread fixes for parsing, warnings, and portability

[Copilot]

This updates benchmark.sh to resolve all comments in the referenced PR review thread, applying the requested behavior changes directly. The fixes are scoped to portability, timing-value parsing, warning criteria, and summary wording.

• Portable temp file creation

  • Replaced bare mktemp with a template-backed invocation so the script works on macOS/BSD and Linux variants.

• Timing parse normalization (zero → 0)

  • Updated parse_dtest_output to normalize dtest’s literal zero timing output to numeric 0 for parsed user/sys/avg fields.

• Warning signal narrowed to required fields

  • Changed incomplete-output warning logic to trigger only when required timing fields (user, sys) are missing, avoiding false warnings when optional fields are absent.

• Summary wording aligned with actual computation

  • Updated compare-summary label from “total user ms” to “avg user ms” to match the existing average-based calculation.

# before
RESULTS="$(mktemp)"
if [[ "$parsed" == *"NA"* ]]; then ...

# after
RESULTS="$(mktemp "${TMPDIR:-/tmp}/dds-benchmark.XXXXXX")"
read -r parsed_user parsed_sys _ _ <<<"$parsed"
if [[ "$parsed_user" == "NA" || "$parsed_sys" == "NA" ]]; then ...