Fix minimum benchmark time with multiple threads

[hahnjo]

Closes #2117

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[dmah42]

test failures look legit.. test/user_counters_test.cc:386: expected (double)foo=143.719 to be EQ to 71.8598
test/user_counters_test.cc:386: with tolerance of 0.0718598 (0.1%), but delta was 71.8592 (99.9992%)

[LebedevRI]

I don't have an opinion on #2117 yet,
but i do want to point out that regardless of whether or not this is a fix,
the behaviour needs a test.

[LebedevRI]

Thinking about this more, without casting judgement on the change itself,
i do feel like pointing out that i very much don't believe that the change
should be framed as a bugfix - i'm reasonably sure that the ->Threads() feature
is intended for testing concurrency, i.e. how the code scales with more threads.

I'm not //sure// if change of behaviour can be done.
That being said this has come up before,
and if we were implementing from a clean slate,
i do believe this would be the default behaviour.

At the very least, it would make sense to have a toggle to
go between the current and proposed behaviours.

So i don't know...