Convert decimal module to use PyLongWriter API

[skirpichev]

Second commit (patch v2) has a quick path for int64_t-sized integers.

Benchmark	master	patch	patch2
1<<7	651 ns	658 ns: 1.01x slower	531 ns: 1.22x faster
1<<38	747 ns	700 ns: 1.07x faster	533 ns: 1.40x faster
1<<300	2.35 us	2.32 us: 1.01x faster	not significant
Geometric mean	(ref)	1.02x faster	1.15x faster

Benchmark hidden because not significant (1): 1<<3000

Details

import pyperf
from decimal import Decimal as D

runner = pyperf.Runner()
runner.bench_func('1<<7', int, D(1 << 7))
runner.bench_func('1<<38', int, D(1 << 38))
runner.bench_func('1<<300', int, D(1 << 300))
runner.bench_func('1<<3000', int, D(1 << 3000))