Support VECTOR data type

[bgrainger]

Discussed in #1548, originally posted by @harrison314.

Add support for the VECTOR(n) type, which is available since MariaDB 11.7.

https://mariadb.com/kb/en/vector-overview/

I've solved it with a workaround:

    internal static class MySqlExtensions
    {
        public static MySqlParameter AddWithVectorValue(this MySqlParameterCollection parameters, string name, ReadOnlySpan<float> vector)
        {
            byte[] buffer = new byte[vector.Length * 4];

            for (int i = 0; i < vector.Length; i++)
            {
                BinaryPrimitives.WriteSingleLittleEndian(buffer.AsSpan(i * 4), vector[i]);
            }

            MySqlParameter p = parameters.Add(name, System.Data.DbType.Binary);
            p.Value = buffer;

            return p;
        }
    }
```</div>

[bgrainger]

VECTOR was also added in MySQL 9.0: https://dev.mysql.com/doc/relnotes/mysql/9.0/en/news-9-0-0.html#mysqld-9-0-0-vectors

[bgrainger]

The underlying type should probably be float[] (not ReadOnlySpan<float> as that can't be boxed for MySqlParameter.Value).

Edit: Will use ReadOnlyMemory<float> as per dotnet/runtime#115148 (comment).

Possibly could add GetFieldValue<ReadOnlySpan<float>> to get a view of the row that's been retrieved.

[bgrainger]

The VECTOR data type was added in Connector/NET 8.4.0.

[bgrainger]

Preliminary tests with MariaDB 11.7 indicate that it doesn't use a dedicated on-the-wire type for VECTOR(n), so all results come back as BLOBs. (Possibly waiting on https://jira.mariadb.org/browse/MDEV-35831.) This means that MySqlDataReader.GetValue(n) will return byte[] and users will have to use Buffer.BlockCopy or MemoryMarshal.Cast<> to get a float[].

[harrison314]

Possibly could add GetFieldValue<ReadOnlySpan<float>> to get a view of the row that's been retrieved.

I think it's enough to implement float[], because the array will always be allocated anyway. I used ReadOnlySpan<float> in my helper method because it is a generic type. But it may not be suitable for implementing methods in the MySQL connector.

This means that MySqlDataReader.GetValue(n) will return byte[] and users will have to use Buffer.BlockCopy or MemoryMarshal.Cast<> to get a float[].

MemoryMarshal.Cast<> may not work on all processors due to big and little endian.

[harrison314]

Don't you plan to release at least a preview nuget with vector support in the near future?

[bgrainger]

@harrison314 Since you mentioned big-endian systems, can you review this one commit? c7bb9b3

(I don't have a big-endian system for testing, so it's mostly a guess that that's the correct code.)

[harrison314]

@bgrainger Unfortunately, I also don't have a big-endian system to test.
I looked at the given commit, and tried the opposite direction of the transformation using both BinaryPrimitives and Array.Reverse and they gave the same results. Therefore, I think the given code is fine.