revisit BFieldElement's to / from native endian bytes

[jan-ferdinand]

This PR fixes the following issues with current conversion from “native” endian bytes to base field elements:

• Make the function infallible. Before, if the passed-in slice was of the incorrect length, the function would panic – see .copy_from_slice(). This behavior was not documented.
• Add a reciprocal method to get a “native” endianess byte array from a base field element.
• Test the behavior.

All that said, I think there are potential footguns regarding cross-platform support lingering here. I would argue we should drop “native” endianness in favor of either little or big endian representation, or maybe drop all of that since we already have from_raw_bytes() and .raw_bytes().

[coveralls]

coverage: 97.618% (+0.05%) from 97.573%
when pulling 2748b52 on bfe_ne_bytes
into b6f4ceb on master.

[Sword-Smith]

I totally agree that we shouldn't even have the option of encoding from/to native endianness. I also remarked that in an issue, as it seems we're actually using native endianness at least one place in our code base: #205