[ialexsilva]

Summary

• Add repeatable --server endpoints for one-off resolver comparisons and let an explicitly selected system DNS resolver remain as a baseline when encrypted protocol filters are used.
• Make the default launch pace adapt conservatively to evidence of shared-path congestion, while keeping an explicitly configured --pace fixed.
• Harden resolver ID handling and retry cancellation so sessions, statistics, and partial reports remain accurate.
• Update the application version to 0.7.0 and document the new selection and pacing behavior.

Resolver selection

• Accept temporary UDP, DoT, DoH, DoH/3, and DoQ endpoints without writing them to the user configuration.
• Preserve the explicitly selected system DNS resolver when --protocols excludes UDP, allowing comparisons such as --system --builtin --protocols doh3,doq.
• Assign deterministic unique IDs when distinct endpoints would otherwise generate the same identifier.

Adaptive pacing

• Treat --concurrency strictly as the in-flight query limit.
• Keep the default 20 ms launch interval adaptive, backing off only when timeout evidence spans multiple independent resolver groups.
• Recover gradually after clean traffic and retain pacing adjustments in events and JSON reports.
• Keep an explicitly supplied --pace fixed and disable adaptive jitter for that mode.

Reliability

• Apply admission control consistently to triage, benchmark attempts, and retries.
• Preserve the last real query result when cancellation prevents a retry from starting.
• Keep persistent sessions isolated by resolver ID and avoid fabricated answered samples in interrupted reports.

Testing

• go test ./...
• go vet ./...
• go test -race ./internal/bench ./internal/serverlist ./cmd/dnsbench