utils.encoding.auto_decode() was broken when decoding Big Endian BOM
byte-strings on Little Endian or vice versa.

The TestEncoding.test_auto_decode_utf_16_le test was failing on Big Endian
systems, such as Fedora's s390x builders. A similar test, but with BE BOM
test_auto_decode_utf_16_be was added in order to reproduce this on a Little
Endian system (which is much easier to come by).

A regression test was added to check that all listed encodings in
utils.encoding.BOMS are valid.

Fixes https://github.com/pypa/pip/issues/6054