Fix metric() rounding carry stopping at quetta

[Vinayak19112003]

Closes #423.

The rounding-carry absorption in metric() was guarded by exponent < 30, so it stopped at quetta (10^30). Values rounding up kept too many digits, and values rounding past quetta were never moved to scientific notation:

>>> metric(9.996e30)
'10.00 Q'   # 4 significant digits, should be 3
>>> metric(9.996e32)
'1000 Q'    # should fall back to scientific

The guard is removed; after the carry, if the exponent reaches 33 (past the largest SI prefix), the function falls back to scientific() as it does for inputs that start there:

>>> metric(9.996e30)
'10.0 Q'
>>> metric(9.996e31)
'100 Q'
>>> metric(9.996e32)
'1.00 x 10\u00b3\u00b3'

Tests: new test_metric_rounding_carry_at_quetta (5 cases) — fails on main, passes with the fix. Full tests/test_number.py: 246 passed. Ruff check and format clean.

[Wkhanex]

I checked the quetta carry boundary in both signs: 9.996e30 → 10.0 Q, 9.996e31 → 100 Q, and 9.996e32 → 1.00 x 10³³, with the corresponding negative outputs. All six passed at this head. Keeping the original value for the scientific fallback avoids formatting the already-rescaled mantissa as if it were the original number. Consider adding negative cases alongside the new positive regressions. I ran these checks on Python 3.12 with English locale hooks, not a full package suite.