SQLAlchemy 1.x and 2.x compatibility: Use explicit transactions, remove sqlalchemy version constraint

[RobinL]

Closes #1785 and #1906 and #1786

It turns out that the UDF registration only works correctly in sqlalchemy 2.0 if you use begin, which starts a transaction explicitly. otherwise subsequent function registrations can't always 'see' previous ones.

fetchall() no longer works in sqlalchemy 2.0.0 see **here but luckily .mappings().all() seems to work fine

I have checked, and the following work:

sqlalchemy.__version__='1.4.49'
pd.__version__='1.5.0'

sqlalchemy.__version__='2.0.25'
pd.__version__='1.5.0'

sqlalchemy.__version__='2.0.25'
pd.__version__='2.2.0'

The only combination that doesn't is:

sqlalchemy.__version__='1.4.49'
pd.__version__='2.2.0'

AttributeError: 'Engine' object has no attribute 'cursor'

But that's fine because #1906 (comment)

there was no version pin for [sqlalchemy] in the pandas 1.x dependencies so you could easily end up installing pandas 1.x and SQLAlchemy 2.x,

[ADBond]

This has got to pretty high up the list of lines changed to issues closed ratio!

Just a couple of things:

• looks like there is still one more reference to .fetchall() in _table_exists_in_database here
• would it be possible to update the changelog?

Happy for you to merge when ready

[ADBond]

Also might be useful to fold these changes into splink4_dev relatively soon while it's fresh in the mind, so that the relevant changes get propagated to their correct new homes

[RobinL]

This has got to pretty high up the list of lines changed to issues closed ratio!

Just a couple of things:

• looks like there is still one more reference to .fetchall() in _table_exists_in_database here
• would it be possible to update the changelog?

Happy for you to merge when ready

Good spots, fixed