Return NaN/Inf unchanged from naturalsize()

[AbhiramMandala]

naturalsize() raised ValueError for math.nan and produced malformed output like 'inf QB' for math.inf, unlike other numeric humanizers that consistently return 'NaN', '+Inf', or '-Inf' for non-finite input via number._format_not_finite().

Add the same isfinite() short-circuit to naturalsize(), reusing number._format_not_finite() for consistent formatting, plus docstring examples and a dedicated parametrized test.

Changes proposed in this pull request:

• Short-circuit naturalsize() with isfinite(bytes_) immediately after bytes_ = float(value), before any exponent/log math runs on the value.
• Reuse number._format_not_finite() for the non-finite return value instead of duplicating nan/inf string formatting, keeping naturalsize() consistent with intcomma, intword, apnumber, fractional, scientific, clamp, metric, and naturaldelta.
• Add three pycon doctest examples (naturalsize(float("nan")), naturalsize(float("inf")), naturalsize(float("-inf"))) to the function's docstring.
• Add a dedicated test_naturalsize_not_finite parametrized test covering math.nan, math.inf, and -math.inf across the default, binary=True, and gnu=True modes.

[AbhiramMandala]

Hi @hugovk, when you get a chance, could you please review this PR? I’d appreciate any feedback. Thanks!