gh-72902: improve Fraction(str) speed

[eendebakpt]

Improve fractions startup time by improving the regexp and using m.groups(). Added tests have been cherry-picked from #133994 by @skirpichev

Benchmark

Benchmark	ref	patch
Fraction('123')	1.09 us	902 ns: 1.21x faster
Fraction('1/3')	1.12 us	946 ns: 1.18x faster
Fraction('1.2e-3')	1.66 us	1.29 us: 1.28x faster
Fraction('-.2')	1.38 us	1.05 us: 1.31x faster
Geometric mean	(ref)	1.24x faster

with script

import pyperf
from fractions import Fraction as F

runner = pyperf.Runner()
s = 'Fraction'
for v in ["123", "1/3", "1.2e-3", "-.2"]:
    r = s + '(' + repr(v) + ')'
    runner.bench_func(r, F, v)

Detailed changes to the regexp

1. Possessive quantifiers (*+, ++, ?+) — match greedily, never backtrack (digits/underscores can't overlap) 2. `[eE]` instead of `re.IGNORECASE` — only `E` needed case folding 3. `\d*|\d+(_\d+)*` to `(?:\d++(?:_\d++)*+)?+` — same language, no alternation, no backtracking. 4. `/denom` branch unconditional — the other alternation branch already matches empty, so the outer ? was redundant. 5. Drop unnamed capture groups around `(_\d+)*`

(modifications found by Claude, removed a few changes to keep diff shorter)

• Issue: Fractions instantiation revisited #72902

[skirpichev]

PR description looks misleading for me. I guess it's a replacement for #133994.