Replace ExecutionList with linked list of RunnableExecutionPair

[stepancheg]

Patch does CAS loop and XCHG instead of synchronized which should
be faster.

Patch also reduces per-future memory consumption.

[slandelle]

Patch does CAS loop and XCHG instead of synchronized which should be faster.

Do you have some numbers please? This code is forked from Guava, so before introducing code that will have to be maintained here, I'd rather be sure it's worth it.

[stepancheg]

Well, it is easy to create a benchmark that proves anything.

I've created a simple benchmark:

https://gist.github.com/stepancheg/2f3236cb5428f215eb2d92e0fafed4f2

It prints:

19158
18589
19825
18905
19007
17067
18801
16763

on current master, and prints

8200
6686
6798
6620
7832
7202
7713
6623
6706

on patched version. Less is better. It is important for benchmark that addListener and runListeners are executed on different threads, otherwise JVM uses lock elision and original version is slightly faster.

However, the most important reason for me is better memory utilization in patched version: there is no ExecutionList object per future.

[slandelle]

Awesome work, thanks!