[Context]: add Ordered/Unordered result ordering control

[rian-be]

Summary

Add Ordered vs Unordered result ordering control for map-style batch operations.

Goal

Let consumers choose, per call: stable predictable ordering (results[5] ≡ items[5], default) or lower-latency streaming order (Unordered).

Reasoning & Build

Ordered (default)

• Predictability; higher memory cost (results buffered to retain index order).

Unordered (opt-in)

• Lower latency ("streaming"); results returned/yielded as soon as they finish.

• Snapshot-internal use: DB insert where order is irrelevant.

• Ordered and Unordered modes surfaced on the relevant batch/map primitives.

• Unordered streams results as they complete (lowest latency).

• Ordered buffers to preserve input index mapping.

• Unit tests for both modes' contract + memory tradeoff note.

Design Expectations

• Ordered is the default; Unordered is clearly opt-in.
• Composes with MapParallel (chunking ordering) and map APIs.

Acceptance Criteria

• Default path preserves input order.
• Unordered path returns as soon as items finish.
• Ordering choice has no impact on context isolation.

Non-Goals

• No reordering semantics beyond index stability / streaming.