gh-119396: Optimize unicode_decode_utf8_writer()

[vstinner]

Take the ascii_decode() fast-path even if dest is not aligned on size_t bytes.

• Issue: Optimize _PyUnicodeWriter implementation #119396

[vstinner]

Benchmark (C part, _testcapimodule.c):

static PyObject *
bench(PyObject *Py_UNUSED(module), PyObject *args)
{
    const char *str;
    Py_ssize_t loops;
    if (!PyArg_ParseTuple(args, "ny", &loops, &str)) {
        return NULL;
    }

    PyTime_t t1, t2;
    (void)PyTime_PerfCounterRaw(&t1);
    for (Py_ssize_t i=0; i < loops; i++) {
        PyObject *obj = PyUnicode_FromFormat("=%s", str);
        if (obj == NULL) {
            return NULL;
        }
        Py_DECREF(obj);
    }
    (void)PyTime_PerfCounterRaw(&t2);
    return PyFloat_FromDouble(PyTime_AsSecondsDouble(t2 - t1));
}

Benchmark (Python part):

import pyperf
import _testcapi
runner = pyperf.Runner()

arg = b'x' * 10
runner.bench_time_func("FromFormat('x'*10)", _testcapi.bench, arg)
arg = b'x' * 1_000
runner.bench_time_func("FromFormat('x'*1_000)", _testcapi.bench, arg)

Result:

+-----------------------+---------+-----------------------+
| Benchmark             | ref     | change                |
+=======================+=========+=======================+
| FromFormat('x'*10)    | 75.9 ns | 68.7 ns: 1.11x faster |
+-----------------------+---------+-----------------------+
| FromFormat('x'*1_000) | 200 ns  | 191 ns: 1.05x faster  |
+-----------------------+---------+-----------------------+