feat: Allow to configure stacktrace for captureMessage calls

[kamilogorek]

Ref #332

Stack trace example:

Removed frame (https://github.com/getsentry/raven-node/compare/message-stacktrace?expand=1#diff-50cfa59973c04321b5da0c6da0fdf4feR337):

[mitsuhiko]

do you have to throw the error to get the stack?

[kamilogorek]

No, it's just more explicit, as we wrap it in try clause. Also this is how we have it written in raven-js and I'm trying to write as unified code as possible.

[mitsuhiko]

I'm mostly asking because throwing an exception vs just creating an error should have noticable performance differences. And since message logging is potentially more common than crashes it might be a concern here.

[kamilogorek]

I'm surprised with this, but you're right : o

throw new Error(i)
608.497162ms
575.068622ms
574.888692ms
540.633688ms
540.919783ms
550.634267ms
577.011109ms
541.462129ms
568.272018ms
573.879691ms

avg: 565.1267161000001ms

new Error(i)
410.160736ms
412.978303ms
423.740643ms
434.467241ms
420.66899ms
426.131056ms
430.911071ms
426.683749ms
429.369119ms
432.251688ms

avg: 424.73625960000004ms

var start = process.hrtime()
var avg = 0 
var n = 10
var sample = 100000

console.log('\nthrow new Error(i)')

for (var j = 0; j < n; j++) {
  for (var i = 0; i < sample; i++) {
    try {
      throw new Error(i)
    } catch (e) {}
  }

  end = process.hrtime(start)[1]
  start = process.hrtime()

  avg += end 

  console.log('' + (end/1000000) + 'ms')
}
console.log('\navg: ' + (avg/n/1000000) + 'ms')

console.log('\nnew Error(i)')

avg = 0 

for (var j = 0; j < n; j++) {
  for (var i = 0; i < sample; i++) {
    try {
      new Error(i)
    } catch (e) {}
  }

  end = process.hrtime(start)[1]
  start = process.hrtime()

  avg += end 

  console.log('' + (end/1000000) + 'ms')
}
console.log('\navg: ' + (avg/n/1000000) + 'ms')
```js

[kamilogorek]

tests are failing without throw, in the end it is required ¯_(ツ)_/¯