fix: catch OverflowError in naturalsize() for very large integers

[koteshyelamati]

Problem

naturalsize() calls float(value) unconditionally, which raises OverflowError for very large integers:

>>> import humanize
>>> humanize.naturalsize(10**309)
OverflowError: int too large to convert to float

It also raises a raw ValueError from float() for non-numeric strings, with a confusing message.

Fix

Wrap float(value) in try/except (ValueError, OverflowError) and re-raise with a clear message indicating the argument type.

After fix

>>> humanize.naturalsize(10**309)
ValueError: naturalsize() argument must be a number, not 'int'
>>> humanize.naturalsize("abc")
ValueError: naturalsize() argument must be a number, not 'str'