Track null arguments in order to provide the appropriate type when converting them.

[etorreborre]

This PR fixes an issue occurring when:

• A value None : Option<T> is bound to a query, and that query executed.
• The same query is later executed with a value Some(t: T) : Option<T>.

Since queries are cached by default, the type of value considered for the bound parameter is Option::<i32>::None.
This will raise a Postgres error on the second execution "incorrect binary data format in bind parameter <n>".

The fix consists in storing the value type of null values (using the result of the encode function to determine if the value is null), and then using a None value with the correct type (for example Option::<String>::None to persist a nullable TEXT value).

I am quite new to the sqlx library and there might be a better implementation but this one seems to be quite consistent with the implementation for the MySQL driver.

[etorreborre]

@abonander the failures seem to be related to some recent Postgres changes on main. Do you confirm this?

[etorreborre]

Hi @abonander. Is the latest version I pushed ok?

[etorreborre]

Thanks @abonander!