[SYTEMDS-3902] Transfer scipy compressed matrices to java runtime

[e-strauss]

This commit implements direct transfer support for Scipy sparse matrices (CSR and COO formats) between Python and the Java runtime.

• Added Java utility methods to convert Scipy Compressed Sparse Row (CSR) and Coordinate (COO) byte arrays directly into internal MatrixBlock representations.
• Updated the Python SystemDSContext to support a sparse_data_transfer flag and extended from_numpy to accept scipy.sparse matrices.
• Enhanced the Python converter logic to detect sparse matrices and either use the new optimized sparse transfer or fall back to dense conversion based on configuration.

Benchmark results:

[e-strauss]

@Baunsgaard @mboehm7 https://github.com/apache/systemds/blob/8de93a1f996348c260d48c1c29340d5a88005e6f/src/main/python/systemds/context/systemds_context.py#L772C4-L772C20

the naming "from_numpy" is misleading since we support now scipy compressed matrices as well
shall we rename it to something like "from_py_matrix", "from_array_like"?

We could also use something like "from_pydata" which merges the two separate methods from_numpy and from_pandas to single method, internally we could route to the corresponding handler based on the input instance type.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 71.54%. Comparing base (b394e32) to head (3184b9a).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #2379      +/-   ##
============================================
+ Coverage     71.51%   71.54%   +0.02%     
- Complexity    47441    47464      +23     
============================================
  Files          1539     1539              
  Lines        182605   182631      +26     
  Branches      35916    35919       +3     
============================================
+ Hits         130585   130655      +70     
+ Misses        42028    41972      -56     
- Partials       9992    10004      +12     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Baunsgaard]

@Baunsgaard @mboehm7 https://github.com/apache/systemds/blob/8de93a1f996348c260d48c1c29340d5a88005e6f/src/main/python/systemds/context/systemds_context.py#L772C4-L772C20

the naming "from_numpy" is misleading since we support now scipy compressed matrices as well shall we rename it to something like "from_py_matrix", "from_array_like"?

We could also use something like "from_pydata" which merges the two separate methods from_numpy and from_pandas to single method, internally we could route to the corresponding handler based on the input instance type.

This sounds good to me, unifying the API would be fine.
However,

1. when doing it do not break backwards compatibility.
2. when calling the old methods report a deprecated message to the user via the logging framework.

[e-strauss]

sure, I agree

[Baunsgaard]

There are a few edge cases not covered.