Normalize scientific mantissas after rounding carries

[kokotatan]

Scientific formatting normalizes the mantissa before rounding, so rounding across a power of ten can add an extra leading digit without updating the exponent. For example, format_scientific(Decimal("9.99"), "0.0E0", locale="en_US") currently produces 10.0E0; it should produce 1.0E1 with the pattern's two significant digits.

Normalize a finite mantissa that gains a leading digit during rounding, adjusting the exponent and its localized sign. The regressions cover positive/negative exponents, crossing zero with explicit exponent signs, exponent padding, multiple mandatory leading digits, negative numbers, Arabic numbering symbols, and preservation of non-carry, zero, infinity, and disabled quantization behavior.

The expected formatting follows the LDML scientific notation rules for integer digits and maximum significant digits. Related open PR #1281 changes initial exponent grouping; this fix addresses rounding carry with Babel's current normalization. PR #1286 addresses directional rounding of negative values, which is a separate issue.

Validation:

• Regression tests before the fix: 9 failed, 4 passed.
• Full suite on Python 3.13 after importing CLDR: 7844 passed, 9 skipped, 2 xfailed.
• All 146 number formatting tests passed.
• All pinned pre-commit hooks passed (all files before the final two context cases; changed files rechecked afterward).
• Independently checked all eight Decimal rounding modes and near both Decimal exponent limits (66 checks).

This contribution was prepared with OpenAI Codex assistance and independently reviewed before submission.