feat(sink): add HarnessSink for Harness AI Evals

[sunilgattupalle]

Summary

• Adds HarnessSink — pushes evaluation scores directly into a Harness AI Evals run
• Follows the same credentials-only constructor pattern as LangfuseSink; required context (harness_run_id, harness_dataset_item_id) comes from eval_case.metadata
• Adds examples/harness_sink_example.py showing end-to-end usage

Design

The sink takes only credentials in the constructor (from params or env vars). Two metadata keys are required on each EvalCase:

• harness_run_id — the run to write into (created externally by the user)
• harness_dataset_item_id — the dataset item this case corresponds to

write() POSTs each item immediately. finalize() PATCHes all seen runs to completed.

Test Plan

• Set HARNESS_ACCOUNT_ID, HARNESS_API_KEY, HARNESS_PROJECT_ID env vars
• Create a dataset + eval + run in Harness AI Evals
• Run python examples/harness_sink_example.py with real IDs
• Verify scores appear in the Harness UI under the run

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}