tail: Add support for hex formatting in `-s` duration

[drinkcat]

Similar to #7475, but unlike timeout and sleep (#7669), we do not support second/minute/hour/day specifier.

GNU coreutils supports hex formatting for the parameter -s, a.k.a. --sleep-interval=N:

tail -s 0x1 -f
tail -s 0x1.1 -f
tail -s 0x1.1p-1 -f

Not a huge deal, probably a pretty rare edge case that I doubt anybody cares about, but it seems worth it to use generic float parsing functions in uucore/parser.

Together with #7669, that would allow to remove dependency on fundu.

[eduardorittner]

I'm looking into this, there's a comment where this is handled:

// Advantage of `fundu` over `Duration::(try_)from_secs_f64(source.parse().unwrap())`:       
// * doesn't panic on errors like `Duration::from_secs_f64`
// * no precision loss, rounding errors or other floating point
// * evaluates to `Duration::MAX` if the parsed number would have exceeded                   
//   `DURATION::MAX` or `infinity` was given  

Should we handle durations that are infinity or exceed DURATION::MAX manually?

[drinkcat]

Yeah... from_str in parse_time.rs does handle that already, maybe we just need a add a boolean parameter to not allow suffixes? Instead of replicating the saturating logic here (we don't necessarily need the underflow logic as 0 doesn't have a special meaning in tail, but I don't think it really matters if we sleep for 0 or 1 nanosecond..)

[eduardorittner]

By just using from_str from parse_time.rs naively, only 3 tests fail:

1. One which exceeds float::MAX: 1.0e100000
2. A single point: .
3. Seconds unit: 1.0s

[eduardorittner]

We can solve the first test by checking that ebd < float64::MAX and returning error on that and the third test by adding a bool to from_str to disallow suffixes, but I'm not sure what do to with the second one.

[eduardorittner]

The check for ebd < float64::MAX would need to be inside from_str which doesn't seem right since from_str returns overflows as Ok, and changing that would probably break all the other callers. Maybe we could add a try_from_str which returns error on overflow?

[drinkcat]

By just using from_str from parse_time.rs naively, only 3 tests fail:

1. One which exceeds float::MAX: 1.0e100000

There's some logic here https://github.com/uutils/coreutils/blob/main/src/uucore/src/lib/features/parser/num_parser.rs#L200 that we could extract that would give us this for free.

2. A single point: .

Wait what? coreutils does not accept that right? But... we do?!

That's yet another bug in our parsing function... Nice catch ,-) #7684

3. Seconds unit: `1.0s

Will be easy with #7675.

[eduardorittner]

Nice! Should I wait on these changes to open a PR?

[drinkcat]

Yeah, maybe best. I think #7675 is close to merging, and I'll work on #7684 (hopefully today). No plan for your point 1 above though.