Use workflow branch 24.08 again

[KyleFromNVIDIA]

After updating everything to CUDA 12.5.1, use shared-workflows@branch-24.08 again.

Contributes to rapidsai/build-planning#73

[jakirkham]

One test was slightly off from its tolerance on CI:

____________ test_tsne_distance_metrics_on_sparse_input[cosine-fft] ____________
[gw7] linux -- Python 3.10.14 /opt/conda/envs/test/bin/python

method = 'fft', metric = 'cosine'

    @pytest.mark.parametrize("method", ["fft", "barnes_hut", "exact"])
    @pytest.mark.parametrize(
        "metric", ["l2", "euclidean", "cityblock", "l1", "manhattan", "cosine"]
    )
    def test_tsne_distance_metrics_on_sparse_input(method, metric):
    
        data, labels = make_blobs(
            n_samples=1000, n_features=64, centers=5, random_state=42
        )
        data_sparse = scipy.sparse.csr_matrix(data)
    
        cuml_tsne = TSNE(
            n_components=2,
            random_state=1,
            n_neighbors=DEFAULT_N_NEIGHBORS,
            method=method,
            learning_rate_method="none",
            min_grad_norm=1e-12,
            perplexity=DEFAULT_PERPLEXITY,
            metric=metric,
        )
    
        if method == "fft":
            sk_tsne = skTSNE(
                n_components=2,
                random_state=1,
                min_grad_norm=1e-12,
                method="barnes_hut",
                perplexity=DEFAULT_PERPLEXITY,
                metric=metric,
                init="random",
            )
    
        else:
            sk_tsne = skTSNE(
                n_components=2,
                random_state=1,
                min_grad_norm=1e-12,
                method=method,
                perplexity=DEFAULT_PERPLEXITY,
                metric=metric,
                init="random",
            )
    
        cuml_embedding = cuml_tsne.fit_transform(data_sparse)
        nans = np.sum(np.isnan(cuml_embedding))
        sk_embedding = sk_tsne.fit_transform(data_sparse)
        cu_trust = trustworthiness(data, cuml_embedding, metric=metric)
        sk_trust = trustworthiness(data, sk_embedding, metric=metric)
    
>       assert cu_trust > 0.85
E       assert 0.8022727822580645 > 0.85

Restarted it

In any event this is unrelated to this change

[jakirkham]

Looks like we got passed that point in the restart

[jakirkham]

/merge

[KyleFromNVIDIA]

/merge

[jakirkham]

Thanks all! 🙏