Improve performance of encoding

[taiki-e]

This change makes encoding about 7x faster than the current implementation, on my machine.

Before:

hex_encode              time:   [77.149 us 77.581 us 78.015 us]

After:

hex_encode              time:   [10.214 us 10.270 us 10.336 us]                        

Benchmarks per commit

Current main branch (aa8f300)

hex_encode              time:   [77.149 us 77.581 us 78.015 us]

First commit of this PR: Adjust #[inline] on encoding functions (97abf03)

hex_encode              time:   [37.251 us 37.412 us 37.583 us]                        
                        change: [-52.047% -51.748% -51.412%] (p = 0.00 < 0.05)
                        Performance has improved.

Second commit of this PR: Use encode_to_slice in encode (0129182)

hex_encode              time:   [27.029 us 27.230 us 27.476 us]                        
                        change: [-27.711% -27.244% -26.782%] (p = 0.00 < 0.05)
                        Performance has improved.

Third commit of this PR: Use chunks_exact_mut instead of generate_iter (19b333d)

hex_encode              time:   [10.214 us 10.270 us 10.336 us]                        
                        change: [-62.198% -61.839% -61.444%] (p = 0.00 < 0.05)
                        Performance has improved.

This also adds encode_to_slice_upper, which is needed to improve the performance of encode_upper. (fixes #45).

[taiki-e]

When using from_utf8_unckecked here, it can improve performance by about 8%. (It is safe because we emit only hex characters.)
However, I didn't apply that change because I don't know the policy regarding the unsafe code in this crate.
If It is okay with using unsafe code, I'll add that change.

hex_encode              time:   [9.4555 us 9.4828 us 9.5106 us]                        
                        change: [-9.2447% -8.3200% -7.4582%] (p = 0.00 < 0.05)
                        Performance has improved.

[cgwalters]

Some overlap with #66 here I think. In particular I suspect the combination of ExactSizeIterator will allow us to collect into a String efficiently without indirecting through Vec.

[yjhmelody]

Nice PR! But the owner of the repo seems to be inactive.