gh-129813, PEP 782: Add PyBytesWriter.overallocate

[vstinner]

• Disable overallocation in _PyBytesWriter_CreateByteArray().
• Disable overallocation in _PyBytes_FormatEx() at the last write.

• Issue: [C API] PEP 782: Add PyBytesWriter API #129813

[vstinner]

Disable overallocation in _PyBytes_FormatEx() at the last write.

Benchmark: python -m pyperf timeit -s 's=(b"a" * 254 + b"%s"); arg=b"b"*100' 's % (arg,)'.

Result: Mean +- std dev: [ref] 165 ns +- 4 ns -> [change] 147 ns +- 1 ns: 1.12x faster. It's 18 ns faster because we avoid a realloc().

[vstinner]

@serhiy-storchaka: This change reintroduces the overallocate=0 micro-optimization in the PyBytesWriter API, but the API to disable overallocation is only available in the internal C API.