Fix intcomma overflow for large Python integers

[Yananix]

Calling intcomma(10**400 + 123) with the default ndigits=None raises OverflowError because the non-finite check converts the integer to float. Python can format this integer directly with thousands separators, so the conversion imposes an unnecessary floating-point range limit.

Changes proposed in this pull request:

• Skip the float validation for values that are already Python integers. Integers are finite, and the existing integer formatting path preserves all digits.
• Add regression cases for positive and negative integers beyond the float range, checking the complete grouped output including the final 123.

Validation on Python 3.12:

• Both regression cases fail with OverflowError before the fix and pass after it.
• Compiled the translation catalogs with scripts/generate-translation-binaries.sh, then ran python -m pytest -q --benchmark-disable: 800 passed, no skips.
• Ruff, Black, and git diff --check passed for the changed files.

This patch is scoped to integer inputs with default precision. It does not change numeric-string parsing or explicit decimal-place formatting.