fix: preserve empty dict handler output

[chenlichao]

Summary

• Stop dropping an empty dict returned by a serverless handler from the normalized job result.
• Add regression coverage for local /runsync and /status simulation paths returning output: {}.

Why this helps

• Fixes Returning empty object from handler raises KeyError exception within framework #459: handlers that intentionally return {} should complete successfully instead of causing local FastAPI simulation to raise KeyError: 'output'.

Test Plan

• pytest tests/test_serverless/test_modules/test_fastapi.py -q --no-cov (8 passed)

Note: a targeted run without --no-cov also had all 8 tests pass, but failed the repository-wide coverage threshold because only one test file was selected.

[capy-ai]

Capy auto-review is paused for this organization because the usage-cycle auto-review limit has been reached. Increase the limit or turn it off in billing settings to resume automatic reviews.

[promptless]

Promptless prepared a documentation update related to this change.

Triggered by runpod-python PR #535

Added a "Handler return values" section to the Serverless handler functions docs, explaining how a handler's return value maps to the job output field — dictionaries pass through directly, other types are wrapped, and an empty dictionary is a valid return that completes as COMPLETED with output: {}.

Review: Document Serverless handler return values

[chenlichao]

bumping this for a first review — one-line fix preserving empty dict handler output, no behavioral regression intended.

[chenlichao]

Weekly bump — still waiting on a first review for this one-line fix preserving empty-dict handler output. No behavioral regression, happy to adjust.

[chenlichao]

Weekly bump — one-line fix preserving empty-dict handler output, with its test; still waiting on a first review.

[chenlichao]

Friendly ping for a review — preserves empty-dict handler output instead of treating it as no result; tests included.

[chenlichao]

Updated the branch: merged current main (70f2de42) in — run_job moved under #570 (prestart hooks) and tests/.../test_job.py changed under #538 (volume cache), so the branch had gone stale/conflicting against main.

While resolving, I also tightened the fix itself. Directly deleting the empty-output pop changes results where popping error/refresh_worker empties the dict: a handler returning {"error": "test"} would become {"output": {}, "error": "test"} instead of the expected {"error": "test"}, and main's test_job_with_errors fails on exactly that (1 failed, 36 passed when I reverted to the delete-only version). The fix now tracks whether the handler itself returned {} and keeps output only in that case, so:

• handler returns {} → {"output": {}} preserved (the point of this PR, covered by the new test_fastapi.py cases)
• handler returns {"error": ...} / {"refresh_worker": true} → results unchanged ({"error": "test"}, {"stopPod": true}; the latter assertion was strengthened to check the exact dict)

Net diff vs main: rp_job.py +3/-1, plus the test_fastapi.py additions and one strengthened assertion in test_job.py.

Tests: full suite via make test — 736 passed, coverage gate green.