Simplify nextFreeRelSlot in ImmDB logic

For the file-backed index, the relatively complicated `nextFreeRelSlot` logic avoids an extra read from disk (although: we're imminently writing to the same page, so does read+write necessarily cost much more than just the write? I don't recall.)

But with the cache, the necessary info is already in memory: `nextFreeRelSlot = length currentChunkOffsets - 1`.

This could alternatively be added as an assertion, perhaps.

Start by reading the ImmDB logic around `nextFreeRelSlot`, comparing the file-backed index path with the cache path and how `currentChunkOffsets` is maintained. Determine whether the cached length can replace the existing calculation or serve as an assertion, then verify that both paths retain the same slot behavior.

Written by the indexing model from the issue text.