Fix Timeout overflow issue caused by negative duration

[xiao66xiang]

Some old sandboxes were not GC'ed because Duration d became negative and caused Timeout::in(d) to overflow. Hence we got a big number for removalTime and those old sandboxes were never sent to remove.

Jira: https://issues.apache.org/jira/browse/MESOS-10232

[bmahler]

While you're here, would you mind updating it to grab a single clock now value at the beginning? The double call makes it tricky to reason about the overflow logic holding true on the second calls as well.

[bmahler]

One more thing, we have to have PRs split across changes to stout/libprocess/mesos, so you'd need one patch for the timeout header and one for the mesos test.

[bmahler]

Looks good, will you be sending a separate PR for the mesos tests?

Can you update the commit title / description to reflect the timeout library issue and not the mesos related issue? We treat libprocess / stout as separate libraries so try to avoid talking about mesos in their commits.

[xiao66xiang]

Yes I'll send a separate PR for the mesos tests.

[mengzhu28]

LGTM. @bmahler let's merge this?

@xiao66xiang please add the mesos test in a separate PR. thanks!

[cf-natali]

@xiao66xiang Thanks!