The main change here is having the `sparrow-execution` tests compile
from a logical plan, created within a session, and execute things.

Changes Include

- Fix naming `field_ref` vs. `fieldref` (matches Python)
- Tweak pipeline scheduling so that only transforms are grouped together.
- Tweak pipelines so that each may output to multiple consumers
- Introduces a `SourceTasks` so we can poll all inputs on a single
  tokio task.
- Fixes to how the scheduler detects "completion" (empty queue isn't
  sufficient).

TODO

- [x] Actually execute the compiled query
- [x] Connect to sources
- [x] Move compilation / execution into session.
- [x] Look at compiling out `tracing::trace` level statements.
- [x] Look at running the query in the sessions tokio runtime
- [ ] Move logical expr building to session (out of sparrow-logical)
- [ ] Replace `ScalarValue` with arrows scalar? Literal too?
- [ ] Look at suspending workers when queue is empty / waking on push
- [ ] Look at a trait for logical functions (type-checking, compilation)