MVStore: use a mark & sweep GC algorithm instead of reference counting, to ensure used chunks are never overwrite, even if the reference counting algorithm does not work properly.

MVStore: use a mark & sweep GC algorithm instead of reference counting, to ensure used chunks are never overwrite, even if the reference counting algorithm does not work properly.

In the multi-threaded mode, database metadata operations did sometimes not work if the schema was changed at the same time (for example, if tables were dropped).

Some CLOB and BLOB values could no longer be read when the original row was removed (even when using the MVCC mode).

Some CLOB and BLOB values could no longer be read when the original row was removed (even when using the MVCC mode).

Some CLOB and BLOB values could no longer be read when the original row was removed (even when using the MVCC mode).

Some CLOB and BLOB values could no longer be read when the original row was removed (even when using the MVCC mode).

In the multi-threaded mode, updating the column selectivity ("analyze") in the background sometimes did not work.

In the multi-threaded mode, database metadata operations did sometimes not work if the schema was changed at the same time (for example, if tables were dropped).

In the multi-threaded mode, updating the column selectivity ("analyze") in the background sometimes did not work.

Fix bug which could generate a NegativeArraySizeException when performing large (>40M) row union operations (v2)

Fix bug which could generate a NegativeArraySizeException when performing large (>40M) row union operations