Cursor/Offset should be replaced

[jjcfrancisco]

The cursor/offset parameter across FastPFOR-rs should be replaced by Bytes::Buf from the bytes crate. The main reason for such change is because the Rust MLT decoder tracks the position internally (also uses Bytes::Buf) - we could jank something but it's not ideal.

[nyurik]

I am really uncertain if this is still needed. The new MLT decoder focuses on slices &[u8], so when decoding, the API might be done in two ways:

• fn decode(input: &[u8], output: &mut [u8]) -> Result<(&[u8], usize), &[u8]>
• fn decode(input: &[u8]) -> Result<(&[u8], Vec<u8>), &[u8]>

In both cases, input buffer is a readonly slice, which gets returned on success as the first param, or as the only value in error. The output could be written directly into the output buffer, returning how many bytes have been written as the second result param. Or it could be dynamically allocated. For the first variant, it may be useful to use MaybeUninit

cc: @CommanderStorm

[CommanderStorm]

I have no clue why, but #47 is 20%-30% faster in the benchmarks that I added.

I honestly have no clue why this is the case since this does not make immediate sense to me.
They must do something that the compiler can optimise much better.

Will need to investigate further, this seems strange at best.