Make Serde document iteration incremental

[jskoiz]

Summary

• Make the public Deserializer consume DocumentStream one parsed document at a time.
• Preserve borrowed string/slice lifetimes, zero-based document indexes, deferred errors, empty-stream null behavior, and existing batch Vec<T> APIs.
• Update the event-backed Serde workpad to describe the current iterator behavior.

Fixes #80

Verification

Local proof:

• cargo fmt --all --check
• git diff --check
• cargo test --locked --lib deserializer_parses_documents_when_they_are_requested
• CARGO_BUILD_JOBS=1 cargo test --locked --test serde_value_api (149 passed)
• CARGO_BUILD_JOBS=1 cargo test --locked --test diagnostics (43 passed)
• CARGO_BUILD_JOBS=1 cargo test --locked --test streaming_api (10 passed)

Hosted/merge state:

• Hosted checks have not yet been observed.
• This PR is intentionally draft and has not been merged.

Notes/Risks

• The explicit batch APIs remain backed by their existing collection path and return Vec<T> as before.
• Reader-backed Serde iteration still buffers the bounded input bytes first; parsed documents are then produced incrementally.
• A repository-wide cargo test --locked attempt reached test-binary linking but could not complete because the host filesystem ran out of space (ld: write() failed, errno=28).