Restructure Dispatch Write Functions to Remove Orphan Writes

[tt-asaigal]

process_write_linear, process_write_paged and process_write_packed_large can be made simpler if we write up to the fence or data length (whichever one comes first), rather than writing single pages and accounting for orphan writes.

These functions got more complex starting at a5a94b4, where orphan writes needed to be accounted for upon block completion.

Writing up to the fence would also allow for the prefetch - dispatch pipeline to be better balanced:

For large txns, writing up to the fence would allow larger writes from the dispatcher (if prefetcher is ahead), thus speeding up dispatch.

Conversely, if dispatcher is ahead, the write size would decrease (the rate at which the fence progresses would be lower), slowing down the dispatcher.

[tt-asaigal]

fyi @pgkeller @tt-aho