Flaky performance test

[jodersky]

The "not show bad performance characteristics when object keys' hashCodes collide" test in JsonParserSpec is flaky.

https://github.com/spray/spray-json/blob/0b893f26ed2fd71649389e7eccc7bba6f1507946/src/test/scala/spray/json/JsonParserSpec.scala#L117

In absolute terms, the regularTime is about 3E6 (3ms) on a modern laptop. This time is sufficiently large to make the test mostly pass on a dedicated machine. However, on a shared environment such as a Travis-CI build, it frequently causes the build to fail.

Potential solutions:

• easy: increase the number of keys by an order of magnitude
• future-proof: don't assume a number of keys will give a large enough time, but rather incrementally deduce that number for a given system time and then run the test

[raboof]

I think we should not put too much effort in these tests - let's first increase the number of keys (and/or be more lenient in the expectation) - if that's not enough I'd even be open to doing a couple of runs or removing the test wholesale.