Share DNS timeout budgets across DoH and UDP fallback

[ryanfowler]

Summary

• Keep one TimeoutBudget alive across DoH POST and JSON fallback so RFC8484 retries cannot restart the timeout.
• Carry the same budget through UDP A/AAAA lookups and truncated UDP-to-TCP fallback so the TCP retry uses the remaining time.
• Update DNS inspection to reuse the budgeted DoH/UDP paths and document the invariant in AGENTS.md.
• Add regressions for delayed DoH 415 fallback and truncated UDP fallback to verify elapsed time stays bounded by the original timeout.

Testing

• cargo fmt
• cargo clippy --locked --all-targets --all-features -- -D warnings
• cargo test --all-features