Bugs: numeric parsing and validation (0.28.1) #34

Defects in the numeric value objects of 0.28.1 (each one reproduced). Fixed by the new numeric.py family; every item gets a regression test and is ticked when the fix lands.

• C09 Int.from_str parses through float and silently changes integers above 2**53 ('9007199254740993' → 9007199254740992). (750b31a)
• C10 Int.from_str accepts non-integer strings that round to an integer ('0.99999999999999999' → 1). (750b31a)
• C11 Int.from_str rejects valid large integer strings that the Int constructor accepts ('1' * 400). (750b31a)
• C12 Int.from_str has no input type check: it accepts bool, bytes, int, float, Decimal and IntEnum. (750b31a)
• C13 Int.from_str / Int.from_float swallow subclass validation errors and replace them with misleading messages. (750b31a)
• C16 Bounded int constructors and Int.from_float(inf/nan) report misleading error messages. (cd2c88e)
• X02 Int.from_str / Float.from_str accept whitespace-padded, underscore-grouped, exponent and non-ASCII-digit strings ('1_000', '١٢', '５', '1e3'). (750b31a, 1031aaa)
• X03 Float.from_int silently rounds integers that are not exactly representable (2**53 + 1). (8953619)
• X04 Int accepts integers whose str_value() then crashes with a plain ValueError (int max-str-digits limit). (7a8a34a)
• C14/C46/C54 ~20 tests named *from_str*_with_decimals call from_float with a string, and the zero tests pass an int to from_str: Int.from_str decimal handling is untested. (750b31a)