UTF8MB4_0900_AI_CI STRINGS

The utf8mb4_0900_bin collation does not behave like other binary
collations when it's mixed with non-binary collations. According to
the documentation: "For an operation with operands from the same
character set but that mix a _bin collation and a _ci or _cs
collation, the _bin collation is used." But when mixing
utf8mb4_0900_bin and for example utf8mb4_0900_ai_ci, the _bin
collation is not preferred, and an error about illegal mix of
collations is raised.

This patch makes some changes to the collation aggregation to make
utf8mb4_0900_bin behave more like other binary collations:

1) Mark utf8mb4_0900_bin with the MY_CS_BINSORT flag so that it is
regarded a binary collation, and is preferred when it's mixed with
non-binary utf8mb4 collations.

2) Reject aggregation of two different binary collations for the same
character set. utf8mb4 is the only character set with more than one
binary collation, so it only affects a mix of utf8mb4_bin and
utf8mb4_0900_bin. When these two collations are mixed, and none of
them has explicit derivation, one cannot determine if the padding
behaviour of utf8mb4_bin should be used, or the non-padding behaviour
of utf8mb4_0900_bin. Mixing them is allowed if a third collation with
explicit derivation is included in the mix.

Change-Id: I73960e71fcb768668c9112603c26f613acc9a92d