fix(sqlalchemy): use connection span in after cursor execute - #901

What changed

receive_after_cursor_execute in sqlalchemy.py had two bugs:

1. Guard check was dead code. get_tracer_tuple() returns a tuple, so if not tracer: was always False — the early-return path never ran.

2. Wrong span was closed. get_current_span() reads from the global OTel context, which could be a different span than the one opened in before_cursor_execute. This corrupted the context stack and caused subsequent child spans (e.g. Redis) to be silently dropped.

Fix: close conn.span directly — the span stored on the connection object by before_cursor_execute — instead of reading from global context.

Tests

• test_if_not_tracing: added assert len(spans) == 0
• test_context_restored_after_query: new test asserting the parent span is active again after a SQL query completes