Unsafe inter-thread communication with appliers

[Gerold103]

There is a not protected reading of non-atomic variables in tx thread which are changing in applier threads. The function applier_read_tx_row() changes applier->lag and applier->last_row_time in applier-thread, and then they are accessed in TX thread without any protection: applier_check_sync() touches applier->lag and lbox_pushapplier() touches applier->last_row_time.

There most likely are more of these misusages.

Given the applier's messy state, it could make sense to split it into 2 structs. One lives entirely in applier-threads, another lives in TX thread. Then the mistakes won't be so easily possible.