[jmaeagle99]

What was changed

Adds the external storage runner: the passes that move payloads to and from a storage driver, and the reference detection they depend on. ExternalStorageRunner offloads eligible payloads from a message on the way out and resolves references back to their payloads on the way in.

Why?

External storage (keeping large payloads out of history by offloading them to a user-supplied driver) has to apply the same offload and retrieval steps at every worker and client path that sends or receives payloads: measure against the threshold, group by driver, and swap references for payloads. This runner is that logic in one place, so those paths call a single component instead of each reimplementing it.

Out of scope

Deliberately not in this change, so the runner can land first:

• Fields are visited one at a time. Per-field concurrency is a later change.
• Reference encoding and decoding are not supported and will be added in a later change.
• No ambient namespace on the select/store contexts, and no Nexus storage target kind.
• Nothing builds the concurrency limiter. The runner takes limiters from the caller, and there is no warning when a driver completes an operation without taking a permit.
• No metrics.
• Payloads inside system Nexus envelopes are not offloaded.
• Per-command targets, and the replay and failure-reporting rules, are left to whatever calls the runner.

Checklist

1. How was this tested: unit tests covering both directions over real messages and payload slices, plus cargo +nightly fmt --check, cargo lint, clippy, and cargo test with and without experimental.

2. Any docs updates needed? No