`detect()` raises `ValueError: Unable to avoid copy` when deprecated `fasttext` PyPI package is co-installed with `fasttext-predict`

[asdasde]

Summary

When the deprecated fasttext distribution (e.g. fasttext==0.9.2) is installed in the same
Python environment as fast-langdetect, every call to fast_langdetect.detect(...) fails
with:

ValueError: Unable to avoid copy while creating an array as requested.
If using `np.array(obj, copy=False)` replace it with `np.asarray(obj)` to allow a copy when
needed
(no behavior change in NumPy 1.x).

The failure is silent at install time — pip reports both distributions as successfully
installed — and only surfaces on the first detect() call.

Reproduction

python -m venv /tmp/repro && source /tmp/repro/bin/activate
pip install "fast-langdetect==1.0.0" "fasttext==0.9.2"
python -c "import fast_langdetect; print(fast_langdetect.detect('hello world', model='lite'))"

Output:

Warning : `load_model` does not return WordVectorModel or SupervisedModel any more, ...
Traceback (most recent call last):
  ...
  File ".../fast_langdetect/infer.py", line 343, in detect
    labels, scores = ft_model.predict(normalized_text, k=k, threshold=threshold)
  File ".../fasttext/FastText.py", line 232, in predict
    return labels, np.array(probs, copy=False)
ValueError: Unable to avoid copy while creating an array as requested.

The same reproduction succeeds (returns a normal language prediction) when only
fasttext-predict is installed.

Root cause

1. fast-langdetect>=1.0.0 declares its dependency as fasttext-predict, a maintained,
   NumPy-2-clean fork that exposes only the predict path.
2. Both the deprecated fasttext distribution and fasttext-predict install files into
   the same top-level fasttext/ package directory (__init__.py, FastText.py,
   fasttext_pybind.*.so). Because the distribution names differ, pip treats them as unrelated
   and silently lets whichever is installed last overwrite the other's files. Verified locally:
   pip uninstall fasttext removes the entire fasttext/ directory even though
   fasttext-predict still claims it.
3. The deprecated fasttext.FastText._FastText.predict ends with:

   return labels, np.array(probs, copy=False)

   Under NumPy ≥ 2.0, copy=False raises if a copy can't be avoided; from a Python tuple of
   floats, a copy can't be avoided, so it always raises.
4. fast_langdetect.detect(...) calls ft_model.predict(...) → ValueError.

Why this matters

Many environments have the legacy fasttext installed transitively (a common dep through
gensim plugins, langid wrappers, embedding libs, etc.). Adding fast-langdetect to such an
environment produces a hard runtime failure with no actionable hint pointing at the packaging
conflict — the traceback points into fasttext/FastText.py, which most users don't realize is
the wrong distribution.

Environment

• Python 3.12.x
• fast-langdetect==1.0.0
• fasttext==0.9.2 (deprecated)
• fasttext-predict==0.9.2.4 (also installed; overwritten by fasttext on disk)
• numpy>=2.0
• Linux x86_64