Reject an empty effective processor chain before starting Batch #77

Summary

Batch.Go(ctx, s, procs ...Processor[T]) currently accepts zero processors. With no processors the pipeline reads the source, discards every item, surfaces source errors, and fires Done(). That is a silent-data-discard footgun: forgetting to pass a processor produces a clean, error-free run that did nothing with the data — the same class of issue as a nil source (now ErrNilSource, see #64).

Proposal

Require at least one Processor. Callers who genuinely want to drain/discard a source pass an explicit no-op (&processor.Nil[T]{}, which already exists), so the intent is explicit at the call site instead of implicit.

• Go with zero processors returns a start error (e.g. ErrNoProcessors), consistent with the ErrNilSource / ErrBatchUsed contract added in feat(batch)!: single-use Batch + Go returns (<-chan error, error); inline item IDs #64.
• "Drain and discard" stays possible, just explicit: b.Go(ctx, src, &processor.Nil[T]{}).

Why

• A zero-stage pipeline silently throwing away data is rarely intended; when it is intended, an explicit no-op documents that.
• Symmetric with the start-error contract in feat(batch)!: single-use Batch + Go returns (<-chan error, error); inline item IDs #64 (nil source → ErrNilSource, reused Batch → ErrBatchUsed).

Tradeoffs to weigh

• Breaks variadic ergonomics for the degenerate zero-stage case.
• Breaks graceful handling of dynamically built processor lists that can legitimately be empty — callers would need to guard or substitute processor.Nil.
• Breaks existing tests that assert no-processors is allowed: TestBatch_NoProcessors (batch/batch_test.go) and the empty-processor-slice case in batch/error_handling_test.go.
• Breaking change (v0) → migration note + CHANGELOG entry.

Alternatives considered

• Document current behavior — keep allowing zero processors, add a godoc line that it drains and discards items (source errors still surface). Lower friction, keeps the footgun.
• Leave entirely as-is.

References

• PR feat(batch)!: single-use Batch + Go returns (<-chan error, error); inline item IDs #64 — start-error contract (ErrNilSource, ErrBatchUsed) and the ExecuteBatches nil-source fix that motivated this question.
• Current behavior lives in Batch.doProcessors (batch/batch.go); the no-op no-processors path is exercised by TestBatch_NoProcessors.