gh-139156: Optimize the UTF-7 encoder

[vstinner]

Remove base64SetO and base64WhiteSpace parameters.

• Issue: Use PyBytesWriter in Unicode codecs #139156

[vstinner]

Remove base64SetO and base64WhiteSpace parameters.

These parameters could be set with the old PyUnicode_EncodeUTF7() C API, but this function has been removed:

PyObject *
PyUnicode_EncodeUTF7(const Py_UNICODE *s,
                     Py_ssize_t size,
                     int base64SetO,
                     int base64WhiteSpace,
                     const char *errors)

[vstinner]

Benchmark:

import pyperf
runner = pyperf.Runner()
sizes = (3, 100, 1_000, 10_000)
for size in sizes:
    runner.timeit(f'{size:,} ASCII chars',
        setup=f's="x"*{size}',
        stmt='s.encode("utf7")')

Results:

Benchmark	ref	change
3 ASCII chars	483 ns	473 ns: 1.02x faster
100 ASCII chars	837 ns	772 ns: 1.08x faster
1,000 ASCII chars	2.93 us	2.48 us: 1.18x faster
10,000 ASCII chars	23.2 us	18.6 us: 1.24x faster
Geometric mean	(ref)	1.13x faster

Note: there is no performance impact on UCS-2 and UCS-4 characters.

[vstinner]

cc @serhiy-storchaka

[vstinner]

Benchmark:

import pyperf
runner = pyperf.Runner()
sizes = (1, 100, 10_000)
for size in sizes:
    runner.timeit(f'{size:,} ASCII chars',
        setup=f's="x"*{size}',
        stmt='s.encode("utf7")')

Results comparing the 3.14 branch to the main branch:

Benchmark	314	main
1 ASCII chars	496 ns	438 ns: 1.13x faster
100 ASCII chars	858 ns	812 ns: 1.06x faster
10,000 ASCII chars	22.8 us	18.4 us: 1.24x faster
Geometric mean	(ref)	1.14x faster

So even with the PyBytesWriter API (PEP 782), the UTF-7 encoder is now 1.14x faster overall on ASCII characters.