Python: Allow replacement scans on pyrelations, move to BoxRenderer and use pending query API for relations

[Mytherin]

This PR makes several improvements to the Python Relational API:

• Move to BoxRenderer for relation objects
• Allows querying relation objects using replacement scans, e.g. this now works:

lineitem = duckdb.query('from lineitem.parquet');
res = duckdb.query('select l_returnflag, sum(l_extendedprice) from lineitem group by all')
print(res)
┌──────────────┬──────────────────────┐
│ l_returnflag │ sum(l_extendedprice) │
│   varchar    │    decimal(38,2)     │
├──────────────┼──────────────────────┤
│ A            │       56586554400.73 │
│ N            │      116422715119.57 │
│ R            │       56568041380.90 │
└──────────────┴──────────────────────┘

• Move to using pending query results, so execution of relations can be interrupted
• Add a cache for the str representation of relation objects

[Mause]

Some nitpicks 😅

[Tishj]

Looks good to me, great piece of functionality and seems like this was relatively low hanging fruit.
That being said it was invisible to me haha

I had one final comment regarding the removal of the GIL unlock to print the progress bar.
Sadly we dont/cant have Jupyter tests, I'm curious if this might break something there

[Tishj]

Do we not want to throw an exception here instead?

[Mytherin]

In general results should not throw exceptions because they might be used in places where those are unexpected.