Simplify tracking implicitly created indexes

[morozov]

Q	A
Type	improvement

I'm trying to clean up the codebase before reworking the logic of managing indexes in the schema.

Currently, Table::$implicitIndexes contains both the index names as keys and the indexes themselves as values. On the one hand, this is redundant (because all indexes are already contained in Table::$_indexes); on the other hand, this is confusing (it is unclear whether Table::$_indexes contains all indexes or only the explicitly created ones).

The proposed fix replaces Table::$implicitIndexes with Table::$implicitIndexNames and addresses both of the above issues.

Additionally, it introduces the PrimaryKeyAlreadyExists exception. This isn't strictly necessary here but also addresses two issues:

1. It breaks a complex if statement that checks for the exceptional condition into two simpler ones.
2. It makes the UX less confusing: unlike regular indexes (which must have a unique name and thus are identified by name), there can be only one primary key, so the name of the offending index in the exception message is irrelevant (furthermore, an index isn't guaranteed to have a name).