Avoid f64 precision loss when parsing integer byte counts

[sayantanmandal1]

Parsing a byte count with a unit suffix loses precision for large integers,
because the suffixed path in from_str always parses the number as f64:

"9007199254740993".parse::<ByteSize>()   // 9007199254740993
"9007199254740993B".parse::<ByteSize>()  // 9007199254740992  <- off by one

f64 only has 53 bits of mantissa, so anything at or above 2^53 gets rounded once
a unit is involved. The no-suffix path already parses straight to u64, so it
stays exact.

When the number has no decimal point this parses it as u64 and multiplies by the
unit factor directly. Decimal inputs like "1.5GiB" still go through f64, and
overflow still saturates like before. Added a test for 2^53 + 1.