Reduce daemon restart delay from 60s to 5s

Summary:
Ref T3557. This delay prevents daemons from hammering overloaded services, but limits how quickly they can process a queue full of bad tasks.

This delay is largely obsolete, because we run so much stuff through taskmasters now, which have more powerful builtin retry delays. I think it's still basically good to keep around, but we can be far less aggressive about throttling.

Reduce the delay from 60s (240 tasks / hour with standard config) to 5s (4,800 tasks / hour with standard config). Even if the failures are load related, pretty much any service should be able to handle 4,800 requests per hour -- the problem is that if we remove it entirely, we might send ~150K+ requests / hour (limited only by how quickly the PHP interpreter can restart), which could plausibly overwhelm unprepared services.

Test Plan: Grepped for constant.

Reviewers: btrahan

Reviewed By: btrahan

CC: aran

Maniphest Tasks: T3557

Differential Revision: https://secure.phabricator.com/D6542