gistfile1.txt

1) Create table "kvs"
2) Create table "kvs_id_idx"

3) For insert:
a) Insert into kvs (rowid, data blob)
b) Insert into kvs_id_idx (key, rowid)

4) For delete:
a) Iterate across kvs_id_idx 
b) Delete kvs where rowid matches 
c) Cursor.Del() on kvs_id_idx

First you need to inject some data for this case to appear.
It seems that with 128 entries of reasonable legnth is fine.
(~75 byte key, ~100 byte value). Key is of format <uuid>/<uuid>.

Then inject some more data, same key/value size. About 128
entries is fine. Then iterate over this second set of data,
and try to delete all the rows. With very high probability
the cursor will not iterate over all 128 entries.

Effectively, I'm simulating multiple indexes by
having a single table store the (rowid, data) tuple, and
all the other tables are just indexes into that table.
Deletes iterate over an index while deleting the true
data rows, however the index scan does not reach all the
expected keys.