arrow

ARROW-3958: [Plasma] Reduce number of IPCs

This PR also removes the client unmap, which is not necessary any more since the introduction of malloc (since there is only few memory mapped files and they typically stay around for the lifetime of the application).

The PR also gets rid of a bunch of code that is not needed any more now (the release buffer, yay!).

Benchmarks:

```
import pyarrow.plasma as plasma

client = plasma.connect("/tmp/plasma", "", 0)

# Put performance

def f():
    for i in range(10000):
        client.put(1)

%timeit f()

# without optimization:

# 1.51 s ± 2.22 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.52 s ± 9.68 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.53 s ± 19 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# with optimizations:

# 1.27 s ± 10.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.31 s ± 8.18 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.31 s ± 17.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# Create/seal performance

def f():
    for i in range(10000):
        object_id = plasma.ObjectID.from_random()
        client.create(object_id, 0)
        client.seal(object_id)

%timeit f()

# without optimizations:

# 571 ms ± 2.28 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 583 ms ± 22.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 588 ms ± 14.9 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# with optimizations:

# 531 ms ± 3.24 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 541 ms ± 9.99 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 542 ms ± 19.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# Get performance

objects = [client.put(1) for i in range(10000)]

def g():
    for i in range(10000):
        client.get(objects[i])

%timeit g()

# without optimizations

# 1.11 s ± 6.17 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.12 s ± 1.49 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 1.19 s ± 24.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# with optimizations

# 776 ms ± 11.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 792 ms ± 3.06 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# 778 ms ± 9.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
```

Author: Philipp Moritz <pcmoritz@gmail.com>
Author: Robert Nishihara <robertnishihara@gmail.com>

Closes #3124 from pcmoritz/plasma-send-fd and squashes the following commits:

f899f459 <Philipp Moritz> Update client.cc
a0384040 <Robert Nishihara> Update _plasma.pyx
af150c14 <Philipp Moritz> comments and fixes
71c4c5c1 <Philipp Moritz> don't close fd twice
0d572823 <Philipp Moritz> linting
f60dcbed <Philipp Moritz> fix tests
502aeda4 <Philipp Moritz> linting
2887b170 <Philipp Moritz> clean up some code
cfff7e32 <Philipp Moritz> lint
e5ccbbac <Philipp Moritz> fixes
5f091993 <Philipp Moritz> introduce method
24beb277 <Philipp Moritz> working version