[GR-60145] Use more efficient impl for math.frexp().

[graalvmbot]

This PR changes math.frexp() to use a more efficient implementation (which is used in test_math.py to calculate fsum() in pure Python). The speedup on a microbenchmark is significant:

# Before
$ graalpy bench_frexp.py
Frexp of 100 numbers: 0.032746 seconds
Frexp of 1000 numbers: 0.020959 seconds
Frexp of 10000 numbers: 0.114700 seconds
Frexp of 100000 numbers: 1.068651 seconds
Frexp of 1000000 numbers: 11.591474 seconds

# After
$ graalpy bench_frexp.py
Frexp of 100 numbers: 0.024994 seconds
Frexp of 1000 numbers: 0.005753 seconds
Frexp of 10000 numbers: 0.012655 seconds
Frexp of 100000 numbers: 0.164891 seconds
Frexp of 1000000 numbers: 2.091557 seconds

# CPython
$ python3.11 bench_frexp.py 
Frexp of 100 numbers: 0.000389 seconds
Frexp of 1000 numbers: 0.004296 seconds
Frexp of 10000 numbers: 0.048125 seconds
Frexp of 100000 numbers: 0.546065 seconds
Frexp of 1000000 numbers: 5.815804 seconds