fix!: the SDK no longer prints to stdout or stderr

[JosteinGj]

What

The SDK printed from library code, and an application embedding it had no way to turn that off.

• process_response wrote every successful response body to stdout, truncated to 2000 characters. Besides the noise, this put response data into whatever captures stdout (container logs, CI output). It truncated at a byte offset, so a multi-byte UTF-8 character at position 2000 (æøå in a name or description) made the call panic after the request had already succeeded.
• The request helpers, datapoint batching, and file download/upload paths eprintln!'d errors. Every one of those errors is also returned to the caller as a ResponseError, so the output only repeated it.
• The datapoint insert printed batch-progress lines on every ingest large enough to need batching.

All of it is removed. The only prints left in src/ are in #[cfg(test)] code.

DataWrapperDeserialization (both the DataWrapper and GraphDataWrapper versions) also lost its non-2xx branch. process_response only calls it for 2xx responses, and the 204 call sites do the same, so that branch never ran. Neither did its error_body fallback. Failed responses become a ResponseError, which is what problem() reads from.

AGENTS.md no longer tells agents to keep the stdout dump. The error_body sentence is corrected to match.

Breaking

• http::process_response is now pub(crate) and no longer takes a path argument (path was used only by the print). It was request plumbing and never meant to be public API.
• Callers who relied on seeing response bodies on stdout while debugging will no longer see them.

Left alone

• error_body on DataWrapper and GraphDataWrapper is now never filled in. The GraphDataWrapper field is pub, so removing it is a separate breaking change.
• A 2xx response whose body won't parse still comes back as Err(ResponseError) carrying the 2xx status. Moving it to 5xx would make is_bufferable() re-send an ingest the server already stored, so this needs a proper decode-failure accessor rather than a different status.

Docs

datahub-sdk-docs: no API signatures change for callers, apart from process_response, which the docs shouldn't reference. If any page suggests running with --nocapture or reading stdout to see response bodies, that advice no longer works.

Tests

• cargo test --all-features: 279 passed, 1 failed, 34 ignored.
• ./run_python_tests.sh: 650 passed, 9 skipped, 1 failed.

Both failures are the file trash lifecycle tests (files::test::tests::file_lifecycle_get_search_update_download_trash_restore, test_files.py::test_get_search_update_download_trash_restore). They were already failing without this change: the tests expect a trashed file's external id to start with DELETED_, and the api no longer adds that prefix. Those tests need updating separately.

🤖 Generated with Claude Code