GH-129149: Add fast paths to four more PyLong_From* functions

[chris-eibl]

Use the macro PYLONG_FROM_INT to implement PyLong_FromInt32() and PyLong_FromInt64(), and the macro PYLONG_FROM_UINT to implement PyLong_FromUInt32() and PyLong_FromUInt64(), because they have a fast path for small and medium-size integers.

before:

PyLong_FromInt32: some small ints
          -5:   228.22 ms
           0:   218.91 ms
         256:   214.52 ms
PyLong_FromInt32: some medium ints
 -1073741823:   313.45 ms
        1000:   196.13 ms
  1073741822:   305.81 ms
PyLong_FromInt32: some bigger ints
 -1073741824:   305.89 ms
  1073741824:   283.47 ms

after:

PyLong_FromInt32: some small ints
          -5:    21.09 ms
           0:    20.80 ms
         256:    20.27 ms
PyLong_FromInt32: some medium ints
 -1073741823:   100.03 ms
        1000:   101.13 ms
  1073741822:    98.36 ms
PyLong_FromInt32: some bigger ints
 -1073741824:   204.82 ms
  1073741824:   205.22 ms

Benchmark done like in #129301 (comment).

• Issue: Missing fast path in PyLong_From*() functions for compact integers #129149

[picnixz]

Are the tests also updated accordingly? (namely do they correctly test all paths?)

[chris-eibl]

Pretty much the same as test_long_as_ssize_t we've discussed in #129301: via round tripping e.g. for in PyLong_FromUInt32

cpython/Modules/_testlimitedcapi/long.c

Lines 772 to 781 in 1a8e574

	static PyObject *
	pylong_asuint32(PyObject *module, PyObject *arg)
	{
	NULLABLE(arg);
	uint32_t value;
	if (PyLong_AsUInt32(arg, &value) < 0) {
	return NULL;
	}
	return PyLong_FromUInt32(value);
	}

and then using check_long_asint, where I've already added more values to cover more cases (especially negative small and medium-size integers): #129301 (comment)

[chris-eibl]

Maybe add even more values? Definitely no performance bottleneck and might gain test coverage. E.g. one more positive and negative small integer?

[picnixz]

Maybe add even more values? Definitely no performance bottleneck and might gain test coverage. E.g. one more positive and negative small integer?

If we're already covering the small, medium and large integers, both in negative and positive ranges, it's fine.

[chris-eibl]

@vstinner are you fine with this? (sorry for the ping, though)

[vstinner]

LGTM