Make it harder for Mean() and StdDev() to overflow

[cassava]

This is my proposal for solving issue #20

[codahale]

This looks great, but the admittedly very touchy tests don’t pass. Feel free to round the stddev and I’ll gladly merge this. Thanks for the patch!

[filipecosta90]

bumping this @cassava .
The tests are still flaky. How do you want to proceed with it? :

:~/go/src/github.com/HdrHistogram/hdrhistogram-go$ make test
GO111MODULE=on go get -t -v ./...
GO111MODULE=on go fmt ./...
GO111MODULE=on go test -count=1 ./...
--- FAIL: TestStdDev (0.01s)
    hdr_test.go:80: StdDev was 288675.1403682712, but expected 288675.1403682715
FAIL
FAIL    github.com/HdrHistogram/hdrhistogram-go 0.153s
FAIL
make: *** [Makefile:35: test] Error 1

[filipecosta90]

@ahothan we can approve this assuming we merge #40 first.

[filipecosta90]

@ahothan I've confirmed that all tests pass with the current master. If we want we can merge it. wdyt?

[ahothan]

@filipecosta90
I was wondering if this could introduces the same kind of issue at the other end of the spectrum of possible values, when total count is extremely high and values are extremely small, but it looks like float64 can handle that in all realistic cases as the smallest number for float64 is ~10**(-308)
So this patch looks good.

[dkropachev]

@filipecosta90 , can you please take a look at it, looks very reasonable.