reduce code redundancy for the `encoding` feature

[pchampin]

A lot of methods are defined twice in the code, once with encoding enabled, and once without it.
Some of these methods have different result types (direct value vs. Result), and this propagate to methods calling them (which must call the former with or without a ?, depending).
(NB: as pointed out in #180, only private method should have a change of signature, but this is orthogonal, I think)

I suggest the following change:

• instead of returning T, private methods in the non-encoding configuration should return std::result::Result<T, std::convert::Infallible>, which after compilation is equivalent to T, but is syntactically is still a result.
• Error should implement From<Infallible> (again, this is merely to please the compiler during parsing: the ‘from method will contain an unreachable!() statement, since Infallible has no value)
• that way, all methods calling the private method can use ?, regardless of the presence/absence of the encoding feature. That way, they don't need two different implementation. The compiler will, however, ignore the ? when the underlying error type is Infallible (i.e. when encoding is disabled).

edited fixed the path of Infallible

[tafia]

This is a breaking change but I think it makes sense yes. (std::convert::Infallible not std::format)

[pchampin]

A breaking change is required also by #180, to stop encoding from being a breaking feature. Kill two birds with one stone :)

And yes, I had made a mistake in the path of Infallible. This is fixed. Thanks for spotting it.