Simpler crud vtab, part 1

[simolus3]

To automatically associate local writes with a transaction id, we use a powersync_crud_ eponymous-only virtual table tracking started transactions and forwarding writes to ps_crud.

But for most writes (the only exception are inserts to insert-only tables), writes actually involve three steps:

1. Creating a ps_crud entry through the vtab.
2. Marking the row as modified in ps_updated_rows.
3. Marking the $local bucket as dirty.

Since the last two steps are almost always the same, this introduces a new powersync_crud virtual table (without a trailing underscore) that behaves almost identical to powersync_crud_, but:

1. Instead of taking a premade JSON string as parameter, it takes the individual fields (e.g. op, id, type, data, old data, metadata).
2. It uses these individual columns to update ps_updated_rows for the correct row / type.
3. It updates $local.

A future step is to make powersync_crud inserts a nop during sync_local, simplifying custom triggers for raw tables.

[simolus3]

I don't think this needs a special migration because the old triggers will continue to work. And since powersync_replace_schema does a diff between the expected and actual triggers after an update, it would just replace them on initialization. But as discussed offline, an empty up migration with a down migration to drop triggers on a core downgrade might be useful because the new powersync_crud vtab wouldn't exist anymore.

There are some failures in the migration tests - mainly because the triggers at older versions look different to the triggers now. I think the only good way to fix these tests is to expect the current trigger definitions everywhere (which is a bit odd when e.g. testing a migration from version 3 to 5, but I don't think the triggers itself should be versioned here).