Preserve zero-valued anonymous tuning metrics

Returning 0 from an objective, or calling tune.report(0), drops the anonymous metric because report tests its truthiness. The default optimizer then raises KeyError: _metric instead of accepting the valid zero loss.

```python
from flaml import tune
tune.run(lambda config: 0.0, config={}, mode="min",
num_samples=1, use_ray=False, verbose=0)
```

Reproduced on Linux / Python 3.12 against main `f9e087c166f2e3feeb1e4b1f57fdd2a53df3ceac`, without Ray. A regression and focused fix are prepared.

This report was prepared with AI assistance.

Start with the tune.run example in the issue and trace how the anonymous objective result reaches tune.report and the default optimizer. Verify that an objective returning 0.0 or tune.report(0) no longer causes KeyError: _metric, and confirm the existing nonzero behavior remains intact.

Written by the indexing model from the issue text.