[Bug Fix] fix eval_recalls for compatibility with NumPy > 1.23

[risingClouds]

Motivation

In NumPy versions prior to 1.23, creating an array from nested lists of varying shapes would result in an object array by default. Starting from NumPy 1.23, such usage raises a ValueError instead. This PR addresses the issue in the eval_recalls function to ensure compatibility with NumPy > 1.23

Modification

Replaced all_ious = np.array(all_ious) with all_ious = np.array(all_ious, dtype=object) in mmdet/evaluation/functional/recall.py, specifically in the eval_recall function (line 111).

BC-breaking (Optional)

No, this modification does not break backward compatibility. The function is adjusted to maintain consistent behavior across NumPy versions.

Checklist

1. ✅ Pre-commit linting tools have been used to ensure code quality.
2. ✅ Unit tests have been updated to verify the fix and ensure compatibility across different NumPy versions.
3. ✅ Tested with downstream projects to ensure no regressions (e.g., MMDet).
4. ✅ Updated the documentation and added comments in the code explaining the change.

[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.

---

risingClouds seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}