Improved resizing and transaction ordering

Database resizing has now been completely internalized and done on the fly within LMDB itself, and does not require throwing any errors or restarting any transactions, which makes it more efficient, easy and deterministic to run transaction callbacks.
There are now new defaults and options for ordering asynchronous transactions and standalone operations.
Synchronous transactions now execute as a child transaction where possible (inside an existing transaction without writemaps or caching).