sqlite3: refactor: recursive traversal; cleanups

[zellyn]

Major change:

• Add visitRawInorder and visitRecordsInorder to btree, which will load and traverse child pages recursively, and use them instead of iterating.

Smaller cleanups:

• Stash the db, not the dbheader in btreeTable, so we can decode child pages.
• Rename load to decodeRecord, and make it take just the payload bytes.
• Create loadCell, a wrapper around parseCell that first seeks to the right place. All callers used to do it themselves.
• Remove unused arg from parseCell, to emphasize that it only parses the cell at the current seek position.
• Remove unused struct type btreeInteriorTable.
• Rename cellInfo.Key to cellInfo.LeftChildPage -- as far as I can tell, no int32 "Key" is mentioned in the data format description -- and don't put sizes in it.
• Make RowID in cellInfo a pointer, since it's optional, but may be zero.

[zellyn]

I realize this is a big change with lots of things mixed together. On reflection, I'm probably too used to our work workflow, where you squash all commits together before merging. Let me know if I should break the changes up into multiple commits.

[sbinet]

nah... it's ok, although I would have perhaps split it in 2: one CL with the recursive btree traversal and another one with the cleanups.

[sbinet]

could you rebase and resolve the conflicts? (apologies for leaving this lingering a bit)

[zellyn]

could you rebase and resolve the conflicts?

Done.

(apologies for leaving this lingering a bit)

On the contrary, you've been remarkably responsive :-)