feat: move hashsolo to sc.pp

[flying-sheep]

Basically just moving the code over and adding an anndata.acc based version of the API on top.

People had issues with our old API since it only supported obs columns:

• https://www.biostars.org/p/9561534/
• https://discourse.scverse.org/t/how-to-install-solo-sc-or-use-hashsolo-from-scvi/1702

The new API of course supports everything! Yay accessors!

• Closes move sc.external.pp.hashsolo into sc.pp #4304
• Tests included or not required because:

• Release notes not necessary because:

[codecov]

Codecov Report

❌ Patch coverage is 91.59664% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.04%. Comparing base (6b5405d) to head (5543fb4).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/scanpy/preprocessing/_hashsolo.py	91.37%	10 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4303      +/-   ##
==========================================
+ Coverage   80.01%   80.04%   +0.03%     
==========================================
  Files         132      133       +1     
  Lines       13397    13403       +6     
==========================================
+ Hits        10719    10729      +10     
+ Misses       2678     2674       -4     

Flag	Coverage Δ
hatch-test.low-vers	78.22% <85.71%> (-0.02%)	⬇️
hatch-test.pre	79.92% <91.59%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/external/pp/_hashsolo.py	100.00% <100.00%> (+11.96%)	⬆️
src/scanpy/preprocessing/__init__.py	100.00% <100.00%> (ø)
src/scanpy/preprocessing/_hashsolo.py	91.37% <91.37%> (ø)

[selmanozleyen]

Wondering which one should be used because it seems inconsistent with

scrublet = deprecated("Import from sc.pp instead")(_scrublet.scrublet)
scrublet_simulate_doublets = deprecated("Import from sc.pp instead")(
    _scrublet.scrublet_simulate_doublets
)

[selmanozleyen]

This works silently for dask and will give a "# FutureWarning: The numpy.column_stack function is not implemented by Dask array."

[selmanozleyen]

the other returns of _calculate_log_likelihoods is never used elsewhere

[selmanozleyen]

we should also mention they should be an integer no? since people can give normalized log easily now

[selmanozleyen]

Maybe a test with 0-len, 1-len, 2-len hashes should be tried out to see if it errors gracefully. Because claude tells me this

"
Silent all-NaN → every cell "Negative" when the derived noise-barcode count hits 0
Both return with no exception: adata.obsm["hashsolo"] is entirely NaN, adata.obs["hashsolo"] is {"Negative": n}, and the only signal is a RuntimeWarning: Mean of empty slice from np.mean(data_sort[:, :-n_barcodes]). One barcode gives IndexError: index -2 is out of bounds for axis 1 with size 1.
"

[selmanozleyen]

In general claude suggested these

test_too_few_noise_barcodes[1-hash] / [2-hashes-default] / [explicit-0]	RuntimeWarning: Mean of empty slice (warning-as-error) instead of the wanted ValueError	#1
test_pre_existing_clusters_missing_label	RuntimeWarning; without warnings-as-error, those rows are [0,0,0] → "Negative"	#2
test_pre_existing_clusters_singleton	RuntimeWarning: divide by zero; otherwise NaN probs	#2
test_pre_existing_clusters for the new ref path

@pytest.mark.parametrize(
    ("n_hashes", "number_of_noise_barcodes"),
    [
        pytest.param(1, None, id="1-hash"),
        pytest.param(2, None, id="2-hashes-default"),
        pytest.param(len(HASHES), 0, id="explicit-0"),
    ],
)
def test_too_few_noise_barcodes(
    adata: AnnData, n_hashes: int, number_of_noise_barcodes: int | None
) -> None:
    """Without a noise barcode, the noise distribution is undefined."""
    with pytest.raises(ValueError, match=r"noise barcodes?"):
        sc.pp.hashsolo(
            adata,
            HASHES[:n_hashes],
            number_of_noise_barcodes=number_of_noise_barcodes,
        )


@pytest.mark.parametrize("kind", ["str", pytest.param("ref", marks=needs.anndata_acc)])
def test_pre_existing_clusters(adata: AnnData, kind: str) -> None:
    """Clustered demultiplexing equals demultiplexing each cluster on its own."""
    adata.obs["cl"] = np.where(np.arange(adata.n_obs) % 2, "a", "b")
    if kind == "ref":
        from anndata.acc import A

        ref = A.obs["cl"]
    else:
        ref = "cl"
    sc.pp.hashsolo(adata, HASHES, pre_existing_clusters=ref)

    np.testing.assert_allclose(adata.obsm["hashsolo"].to_numpy().sum(axis=1), 1)
    for cluster in ("a", "b"):
        sub = adata[adata.obs["cl"] == cluster].copy()
        sc.pp.hashsolo(sub, HASHES)
        np.testing.assert_allclose(
            sub.obsm["hashsolo"].to_numpy(),
            adata.obsm["hashsolo"].loc[sub.obs_names].to_numpy(),
        )


def test_pre_existing_clusters_missing_label(adata: AnnData) -> None:
    """Unlabeled cells must not silently turn into confident negatives."""
    adata.obs["cl"] = pd.Categorical(["a"] * (adata.n_obs - 10) + [None] * 10)
    sc.pp.hashsolo(adata, HASHES, pre_existing_clusters="cl")
    np.testing.assert_allclose(adata.obsm["hashsolo"].to_numpy().sum(axis=1), 1)


def test_pre_existing_clusters_singleton(adata: AnnData) -> None:
    """A one-cell cluster has zero variance; it must not yield NaN probabilities."""
    adata.obs["cl"] = ["a"] * (adata.n_obs - 1) + ["b"]
    sc.pp.hashsolo(adata, HASHES, pre_existing_clusters="cl")
    assert np.isfinite(adata.obsm["hashsolo"].to_numpy()).all()

[selmanozleyen]

Some edge cases inherited from main and one new behaviour with the pd.unique and loc usage instead of the old np.unique. Plus one nitpick on dead return variables. But the rest seems good