Fix ramp-up worker scaling

[hemanth-asirvatham]

Motivation

• The ramp-up logic wasn't actually increasing the active worker pool over time, so concurrency never reached the intended ramp target during get_all_responses runs.

Description

• Added a worker_spawner loop and a helper _desired_worker_count() to gradually spawn new worker tasks as the ramp increases the parallel cap in get_all_responses (file src/gabriel/utils/openai_utils.py).
• Make the spawner call _maybe_emit_ramp_complete() so ramp completion is emitted even when the periodic status_reporter is disabled.
• Initialize an adaptive workers list and start an initial set of workers via _desired_worker_count() instead of creating a static pool upfront.
• Ensure the spawner is cancelled and awaited during shutdown so spawned workers are cleaned up properly.

Testing

• Ran pytest tests/test_parallelization_tuning.py::test_ramp_up_increases_parallelism, which passed.

---

Codex Task

[github-actions]

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 can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}