Share the DNS timeout budget across deliverability lookups - #178

Summary

Apply timeout to all DNS lookups for one email address, as requested in #101. Currently, each fallback lookup can use the full timeout again. For example:

timeout = 10 seconds
MX lookup takes 9 seconds and finds no records
next lookup gets 1 second, instead of another 10

Compute a deadline with time.monotonic() and pass the remaining time to each MX, A, AAAA or TXT lookup using dnspython's lifetime argument. When a resolver is supplied, its lifetime provides the initial timeout. Cached answers can still be returned after the time runs out. Resolver backoff and processing can make the total elapsed time exceed the timeout.

Evidence

• The deliverability tests produce 13 failures and 16 passes on the original code. The revised code passes all 336 offline tests on Linux Python 3.10 through 3.14 and Windows Python 3.14. Typing and lint also pass.
• Tests check the time passed to successive lookups, repeated validations, cached answers after expiry, resolver subclasses and timeout results after a fallback lookup succeeds.

Merge danger

Custom resolve overrides must accept the lifetime keyword argument. Overrides accepting only a domain and record type will fail. The code also stops changing the process-wide default resolver's lifetime. These compatibility changes need maintainer review.

Unverified

Live DNS tests were excluded. Hard cancellation, concurrent changes to the resolver, downstream compatibility, packaging, macOS, other Windows versions, Windows coverage HTML and scanner parser coverage were not verified. Hosted CI has no recorded jobs. The full suite was not run with dnspython 2.0.0 because the test fixture requires its missing CacheKey type; focused cache and resolver probes passed on that version.