RLS with Jinja macro renders SQL Lab use impossible (empty flask session)

### Bug description

We have a base RLS filter applying a clause that looks like `column_name IN {{ rls() }}`. The rls function is a custom Jinja macro that essentially looks in the current flask session for a list and formats it properly so that the final value may look like `column_name IN ('a', 'b', 'c')`:

```python
def rls():
"""
Returns RLS roles formatted for SQL IN clause.
Example: ['admin', "Joe's role"] -> ('admin', 'Joe''s role')
Returns (NULL) if empty.
"""
# session['rls'] filled in during authentication process by interacting with some backend services of ours
values = session.get("rls", [])
if not values:
return "(NULL)"

binds = [bindparam(f"v_{i}", v) for i, v in enumerate(values)]
escaped = [str(b.compile(dialect=postgresql.dialect(), compile_kwargs={"literal_binds": True})) for b in binds]
return f"({', '.join(escaped)})"
```

This works fine in most places, although I've been bitten by https://github.com/apache/superset/issues/33346 (which should really be indicated clearly somewhere in the docs!).

However, if I open a chart (where this RLS is working) in SQL lab using the dropdown menu, Then this `rls()` macro malfunctions (returning me an empty list): enumerating the values in `flask.session` shows that the whole thing is empty in that context, even the default values Superset provides by itself.

This prompts two questions:
- Why is the flask session filled out in all contexts but this one? Is there something particular about the way queries are built in the SQL Lab?
- Why does the RLS even apply in this view? Isn't it supposed _not_ to trigger inside the SQL Lab?

This happens on both 5.0.0 and 6.0.0.

### Screenshots/recordings

_No response_

### Superset version

6.0.0

### Python version

I don't know

### Node version

I don't know

### Browser

Firefox

### Additional context

_No response_

### Checklist

- [x] I have searched Superset docs and Slack and didn't find a solution to my problem.
- [x] I have searched the GitHub issue tracker and didn't find a similar bug report.
- [x] I have checked Superset's logs for errors and if I found a relevant Python stacktrace, I included it here as text in the "additional context" section.