[GR-60145] Use XSum port for math.fsum().

[graalvmbot]

This PR changes math.fsum() to use the port of XSum also used in GraalJS. The speedup on a microbenchmark is significant both with compilation enabled and in interpreter-only mode:

# GraalPy without XSum (with compilation)
$ graalpy bench_fsum.py
Sum of 100 numbers: 0.021997 seconds
Sum of 1000 numbers: 0.005800 seconds
Sum of 10000 numbers: 0.054701 seconds
Sum of 100000 numbers: 0.517120 seconds
Sum of 1000000 numbers: 4.981345 seconds

# GraalPy with XSum (with compilation)
$ graalpy bench_fsum.py
Sum of 100 numbers: 0.025060 seconds
Sum of 1000 numbers: 0.007285 seconds
Sum of 10000 numbers: 0.026717 seconds
Sum of 100000 numbers: 0.140376 seconds
Sum of 1000000 numbers: 1.381729 seconds

# GraalPy without XSum (interpreter only)
$ graalpy bench_fsum.py
Sum of 100 numbers: 0.009417 seconds
Sum of 1000 numbers: 0.027649 seconds
Sum of 10000 numbers: 0.166529 seconds
Sum of 100000 numbers: 1.514038 seconds
Sum of 1000000 numbers: 14.632675 seconds

# GraalPy with XSum (interpreter only)
$ graalpy bench_fsum.py
Sum of 100 numbers: 0.007540 seconds
Sum of 1000 numbers: 0.024484 seconds
Sum of 10000 numbers: 0.101314 seconds
Sum of 100000 numbers: 0.809227 seconds
Sum of 1000000 numbers: 7.568794 seconds

# CPython
$ python3.11 bench_fsum.py
Sum of 100 numbers: 0.001679 seconds
Sum of 1000 numbers: 0.015773 seconds
Sum of 10000 numbers: 0.126783 seconds
Sum of 100000 numbers: 1.238006 seconds
Sum of 1000000 numbers: 12.508949 seconds