gh-102471: convert decimal module to use PyLong_Export API (PEP 757)

[skirpichev]

• Issue: The C-API for Python to C integer conversion is, to be frank, a mess. #102471

For reviewers: in the history included version, where PyLong_Export always set digits.

Benchmark	ref	patch	no-value
Decimal(1<<7)	735 ns	not significant	721 ns: 1.02x faster
Decimal(1<<38)	809 ns	738 ns: 1.10x faster	846 ns: 1.05x slower
Decimal(1<<300)	1.93 us	2.05 us: 1.06x slower	1.99 us: 1.03x slower
Geometric mean	(ref)	1.01x faster	1.01x slower

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

benchmark code

# export_bench.py
import pyperf
from decimal import Decimal as D

runner = pyperf.Runner()
i1, i2, i3, i4 = 1 << 7, 1 << 38, 1 << 300, 1 << 3000
runner.bench_func('Decimal(1<<7)', D, i1)
runner.bench_func('Decimal(1<<38)', D, i2)
runner.bench_func('Decimal(1<<300)', D, i3)
runner.bench_func('Decimal(1<<3000)', D, i4)

[skirpichev]

@vstinner, @serhiy-storchaka, please review.

I removed const's, added assert's, also handling value case now more simple. It seems, difference too small to measure on my system:

Benchmark	ref	patch	only-qset_i64
int(Decimal(1<<7))	620 ns	634 ns: 1.02x slower	634 ns: 1.02x slower
int(Decimal(1<<38))	697 ns	728 ns: 1.04x slower	719 ns: 1.03x slower
int(Decimal(1<<300))	2.02 us	2.04 us: 1.01x slower	2.04 us: 1.01x slower
Geometric mean	(ref)	1.02x slower	1.02x slower

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

(*) only-qset_i64 - current version.

@serhiy-storchaka, if you prefer runtime checks over asserts - I'll cache layout parameters in the decimal module state (though, seems odd for me).

[serhiy-storchaka]

LGTM. Thanks.

[vstinner]

LGTM. I'm a little bit sad that the change makes the code a little bit slower, but I guess that's the price of abstraction.

[vstinner]

Merged, thanks.