implement lock-free semaphore

[pongad]

Previous implementation of Semaphore64 allows two ways to acquire
permits:

1. acquire() - the method will eventually return after it has acquired
   permits.
2. tryAcquire() - the method returns immediately, reporting whether or
   not it have acquired the permits.

Pubsub implements its own work queue and use tryAcquire exclusively.

The problem:
To implement acquire(), we must use a lock
(I do not know of a better way)
and all users must pay for the lock, even if they never use acquire().
In particular, we call notifyAll() when releasing permits;
these calls are quite expensive.

This commit splits Semaphore64 into two implementations,
one exclusively blocks and the other exclusively not.
In this way the non-blocking implementation needs not pay for the cost
the lock.

This reduces CPU usage of pubsub client lib by about 10%.

[codecov-io]

Codecov Report

Merging #407 into master will decrease coverage by 0.01%.
The diff coverage is 85.71%.

@@             Coverage Diff              @@
##             master     #407      +/-   ##
============================================
- Coverage      71.7%   71.69%   -0.02%     
- Complexity      751      753       +2     
============================================
  Files           155      156       +1     
  Lines          3354     3363       +9     
  Branches        261      262       +1     
============================================
+ Hits           2405     2411       +6     
- Misses          847      848       +1     
- Partials        102      104       +2

Impacted Files	Coverage Δ	Complexity Δ
...va/com/google/api/gax/batching/FlowController.java	78.12% <100%> (-0.67%)	17 <0> (-1)
...com/google/api/gax/batching/BlockingSemaphore.java	80.95% <80.95%> (ø)	6 <6> (?)
.../google/api/gax/batching/NonBlockingSemaphore.java	81.25% <81.25%> (ø)	5 <5> (?)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bdcd4d8...433dc8a. Read the comment docs.

[pongad]

I'll hold off on this a bit. Depending on how pubsub simplification goes, this might not add much value. In that case, we might as well not complicate this.

[pongad]

@garrettjonesgoogle Looks like this will be useful after all. PTAL

[pongad]

@garrettjonesgoogle PTAL

[garrettjonesgoogle]

LGTM