gh-158451: Optimize PyUnicode_Join()

[vstinner]

Move tests on the separator outside the loop. Add a loop version for empty separator.

• Issue: C API: Rework and enhance the PyUnicodeWriter implementation #158451

[vstinner]

Benchmark:

import pyperf
runner = pyperf.Runner()
for sep in ('', '.'):
    for list_size in ('10', '10**3', '10**6'):
        runner.timeit(
            f'sep {sep!r}, {list_size} short items',
            setup=f"data=['x']*{list_size}; sep={sep!r}",
            stmt="sep.join(data)")

    for list_size in ('10', '10**2', '10**4'):
        runner.timeit(
            f'sep {sep!r}, {list_size} long items',
            setup=f"data=['x' * 100]*{list_size}; sep={sep!r}",
            stmt="sep.join(data)")

Results on Fedora 44 with CPU isolation:

Benchmark	ref	change
sep '', 10 short items	155 ns	143 ns: 1.08x faster
sep '', 10**3 short items	8.22 us	8.03 us: 1.02x faster
sep '', 10**6 short items	8.58 ms	8.46 ms: 1.01x faster
sep '', 10 long items	240 ns	235 ns: 1.02x faster
sep '', 10**2 long items	1.16 us	1.04 us: 1.12x faster
sep '.', 10 short items	197 ns	181 ns: 1.09x faster
sep '.', 10 long items	282 ns	274 ns: 1.03x faster
sep '.', 10**2 long items	1.38 us	1.39 us: 1.01x slower
Geometric mean	(ref)	1.02x faster

Benchmark hidden because not significant (4): sep '', 10**4 long items, sep '.', 10**3 short items, sep '.', 10**6 short items, sep '.', 10**4 long items

[vstinner]

See also #158689 which optimizes bytes.join().