Before reworking Index similarly to UniqueConstraint and ForeignKeyConstraint, I want to introduce PrimaryKeyConstraint. Currently, they are represented as indexes with the "is primary" attribute set to true, but there's no such thing as "primary index". On all supported platforms, a primary index is implemented as a PRIMARY KEY constraint.

Besides introducing such a class, I want to provide an upgrade path and implement a new API (e.g. Table#getPrimaryKeyConstraint()) which would be state-compatible with the existing API. For instance, a user should be able to create a primary index and then access the primary key constraint.

For that, I want to deprecate all invalid states that an index may currently have. The IndexedColumn class, particularly, will provide some guarantees for indexed column names and their lengths. Also, it will combine the name and the length together, which should be easier to use.

Additionally, even though it currently works correctly, I'm deprecating using nullable columns in a primary index. The reasons are:

1. It's a made-up requirement. Columns in DBAL are non-nullable by default, so someone has to manually to declare it as nullable for that feature to be relevant. Implicitly overriding the definition that the user has explicitly provided looks shady.
2. It mutates the column, while we're moving towards immutable types.