Reduce LfnBuffer size from 24 to 16 bytes on 32-bits sys.

[Conaclos]

• Reduce LfnBuffer size from 24 to 16 bytes on 32-bits sys.
• Reduce LfnBuffer size from 32 to 24 bytes on 64-bits sys.

The reduction comes from the observation that the required buffer doesn't need to exceed the maximum bytes required to represent a LFN. A LFN has at most 255 UTF-16 characters. The buffer holds UTF-8 characters. A UTF-8 character takes at most 3 bytes. So, the buffer doesn't need to exceed 255*3 bytes. Thus, free can be stored in a u16.

To avoid a breaking change, the buffer is sliced in order to retain at most u16::MAX bytes.

Note that the optimization could go beyond, however this could become too complex IMO.

[thejpster]

Thank you for the PR. Looking at this, I think I can just about justify the complexity for the memory saving. But perhaps you could add a function like:

fn free(&self) -> usize {
    usize::from(self.free)
}

That would avoid the need to do so many u16 to usize conversions in the code.

Have you found the eight byte saving to be significant in your application?

[Conaclos]

Have you found the eight byte saving to be significant in your application?

The gain is not significative at an application level. However, I thought it was a good tradeoff between complexity/gain. If you think it is not worth the complexity, feel free to reject it :)