[data] Show external consumer bytes in verbose operator progress log

[justinvyu]

Description

When iterating with ds.iter_batches() or consuming a streaming_split() shard, the consumer's prefetched bytes are charged to the terminal operator's out memory (via set_external_consumer_bytes). Today the progress log shows the combined number without distinguishing the operator's own queues from what the downstream iterator is holding, which makes it hard to tell how much memory the iterator's prefetch is using.

This adds external_consumer=... to the verbose (in=..., out=...) field on the terminal operator's progress line whenever an external consumer is registered:

Before:

split(4, equal=True): Tasks: 0; Actors: 0; Queued blocks: 2 (256.1MiB); Resources: 0.0 CPU, 17.9GiB object store (in=0.0B,out=17.9GiB);

After:

split(4, equal=True): Tasks: 0; Actors: 0; Queued blocks: 2 (256.1MiB); Resources: 0.0 CPU, 17.9GiB object store (in=0.0B,out=17.9GiB,external_consumer=15.2GiB);

The field only appears on the terminal operator (since external consumers attach there) and only when a consumer is registered, so existing logs for pipelines without external consumers are unchanged.

Additional details

We're going to eventually remove this "external consumer tracking" logic in Ray Data, but for now this log is useful for debugging at least internally.

[gemini-code-assist]

Code Review

This pull request updates the ResourceManager to surface external-consumer bytes in the verbose operator usage string of the terminal output operator, allowing users to see how much of the output memory is held by downstream iterators. A new unit test has been added to verify this behavior. There are no review comments, so I have no feedback to provide.