gh-136681: optimize builtin calls if kwargs allowed

[skirpichev]

This is a first part, that should address #136681. With this patch there should be no overhead for function calls if it's signature allows keyword arguments.

Below are micro-benchmarks for math.fmin(), when only positional arguments are allowed (as in the main) vs positional-or-keyword.

The missing part is specialization for 1-arg functions, like cmath.sin(): #136681 (comment)

In the main:

Benchmark	posonly-patch	posorkw-patch
fmin(1.0, 2.0)	169 ns	170 ns: 1.01x slower
fmin(1.0, 2.0) x 2 times	913 ns	890 ns: 1.03x faster
fmin(1.0, 2.0) x 100 times	14.7 us	14.8 us: 1.01x slower
Geometric mean	(ref)	1.00x faster

Benchmark hidden because not significant (1): fmin(1.0, 2.0) x 10 times

With the patch:

Benchmark	posonly-ref	posorkw-ref
fmin(1.0, 2.0)	177 ns	199 ns: 1.12x slower
fmin(1.0, 2.0) x 2 times	937 ns	980 ns: 1.05x slower
fmin(1.0, 2.0) x 10 times	2.08 us	2.25 us: 1.08x slower
fmin(1.0, 2.0) x 100 times	14.8 us	16.5 us: 1.12x slower
Geometric mean	(ref)	1.09x slower

Benchmark code:

import pyperf
from math import fmin

def f(n):
    for _ in range(n):
        fmin(1.0, 2.0)

runner = pyperf.Runner()
runner.bench_func('fmin(1.0, 2.0)', fmin, 1.0, 2.0)
for n in [2, 10, 100]:
    s = f'fmin(1.0, 2.0) x {n:3} times'
    runner.bench_func(s, f, n)

• Issue: Slow functions calls for C-coded functions, accepting positional-or-kwarg argument vs positional-only #136681