With manual transactions, race conditions on `last_insert_rowid()` need to be accounted for

## Race condition scenario with new configuration:

```python
# pseudocode

# We insert the record here!
db.insert(...)

# ! Record adding in different process increments the ROWID by another one

# Let's see the last PK, but now it's been incremented 0-to-many times
print(db.last_pk)
```

## Probably not a fix

```python
db.begin()
db.insert(...)
print(db.get_last_rowid()) # Not necessarily the record
db.commit()
```

Start by tracing the db.insert, db.last_pk, and db.get_last_rowid entry points described in the issue, then review how manual transactions interact with concurrent processes. Reproduce the interleaving shown in the pseudocode and establish the expected primary-key behavior before deciding what a complete fix or test should cover.

Written by the indexing model from the issue text.