Cap lookup time across fallback drivers

[tegos]

Closes #182.

http.timeout is per driver, not per lookup. With the shipped config (IpApi plus four fallbacks) an unreachable provider costs that timeout five times over, so Location::get() in #182 returned false after ~16s with timeout => 3. Reproduced against a socket that accepts but never responds: 14.02s currently, 4.05s with total_timeout => 4.

total_timeout (default null, so nothing changes unless set) gives the whole lookup one budget. Each driver's timeout is capped to what's left, and drivers with no time left aren't called.

I dropped the failure event that was here before to keep this focused on the timeout.

Your call: total_timeout defaults to null for compatibility, which means #182's config still takes ~15s. Happy to default it to a number instead.

[tegos]

Hey @stevebauman, any chance you could take a look at this when you have a moment? No rush.

Heads up that CI hasn't run yet, it needs an approval since it's my first PR here.

[tegos]

Hello! Bumping this one.
If it's too much in one go, I can split it - total timeout in one PR, the LookupFailed event in another.
Or trim it to just whichever half you'd want.

Either way, tell me what shape works and I'll rework it.

[tegos]

Trimmed this down to just the timeout budget. Dropped the failure event, process() keeps the old rescue(). One commit, 180 lines.

CI needs your approval to run.

[stevebauman]

Released in v7.7.0!

[tegos]

Thanks, no problem about delay.

Now we can limit the total time.

With a few fallback drivers, one dead provider made the request wait for all their timeouts.