Let stage components select their own distribution

[ddeboer]

Problem

Currently Stage.run(dataset, distribution) receives a single pre-resolved Distribution and passes it to all selectors and executors. This works when every component in a stage needs the same distribution type (e.g. a SPARQL endpoint), but breaks down when components need different types — for example, a SPARQL Anything stage would need a CSV distribution, not a SPARQL endpoint.

Proposed direction

1. Component-level distribution selection — each selector/executor declares what distribution type it needs and receives it from the resolved context, rather than having a single distribution pushed in from Stage.run()
2. Symmetric selector/executor interface — StageSelector should receive the dataset/context at iteration time (like Executor.execute() does), instead of being fully configured at construction time

Context

This came up during #75, where SparqlSelector takes a bare URL at construction while SparqlConstructExecutor receives the Distribution at execution time — requiring manual FROM clause construction for the selector.

Related: #18 (SPARQL Anything for non-RDF origin data) would benefit from this, since those stages need non-SPARQL distributions.