gh-139156: Use PyBytesWriter in UTF-32 encoder

[vstinner]

Replace PyBytes_FromStringAndSize() and _PyBytes_Resize() with the PyBytesWriter API.

• Issue: Use PyBytesWriter in Unicode codecs #139156

[vstinner]

Benchmark ASCII characters:

import pyperf
runner = pyperf.Runner()
for size in (3, 100, 1000):
    runner.timeit(f'{size:,} chars',
        setup=f's="x"*{size}',
        stmt='s.encode("utf32")')

Result:

Benchmark	ref	pep782
100 chars	58.9 ns	61.5 ns: 1.04x slower
1,000 chars	185 ns	188 ns: 1.02x slower
Geometric mean	(ref)	1.02x slower

Benchmark hidden because not significant (1): 3 chars

[vstinner]

Benchmark UCS-4 characters:

import pyperf
runner = pyperf.Runner()
for size in (3, 100, 1000):
    runner.timeit(f'{size:,} UCS-4 chars',
        setup=f's=chr(0x10ffff) * {size}',
        stmt='s.encode("utf32")')

Result:

Benchmark	ref	pep782
3 UCS-4 chars	54.7 ns	63.9 ns: 1.17x slower
100 UCS-4 chars	99.4 ns	101 ns: 1.02x slower
1,000 UCS-4 chars	546 ns	542 ns: 1.01x faster
Geometric mean	(ref)	1.06x slower

[vstinner]

cc @serhiy-storchaka

[vstinner]

The only significant difference is on "3 UCS-4 chars": 54.7 ns => 63.9 ns: 1.17x slower, +9.2 seconds. That's the cost of the PyBytesWriter API abstraction.

I added a "fast path" for UCS-1 which is the most common cases: it keeps PyBytes_FromStringAndSize(NULL, size). So there is no impact on performance.

[serhiy-storchaka]

LGTM. 👍

[vstinner]

Merged, thanks for the review @serhiy-storchaka.