Raise Invalid for Infinity/NaN input to Number validator

[SAY-5]

Fixes #541.

Number(precision, scale) documents :raises Invalid:, but Number(precision=6, scale=2)("Infinity") (also "NaN"/"-Infinity") leaks a raw TypeError("infinity and NaN have no precision"). Decimal(number) succeeds for these, then as_tuple().exponent is a string rather than an int, so _get_precision_scale fell into the else branch that raised TypeError. The commented-out line there already showed the intended behaviour.

The else now raises Invalid with the same message the InvalidOperation sibling uses, so a non-finite value is rejected like any other unusable number. Added test_number_validation_with_infinity_and_nan mirroring test_number_validation_with_string; it fails before the change and passes after. Full suite: 168 passed, flake8 clean.

This is separate from #539, which handles None/dict/bytes before the Decimal() call and explicitly leaves this non-finite path alone.

[alecthomas]

This has conflicts now that #539 is merged.

[SAY-5]

Rebased onto master. #539 changed the Decimal(...) except clause, but the Infinity/NaN branch this PR targets is the else after the exponent check, which still raised a raw TypeError, so the two changes are complementary. Conflicts are resolved and the branch is a single commit on current master.

[alecthomas]

Thanks!