Add Sparse Vectors Decoding

[devsimiyu]

Description

Adds initial support for Spark ML linalg vectors on the C++ client. Introduces DenseVector and SparseVector classes under src/ml/linalg/vectors/ and instructs Row::get<T>() to decode Spark's vector UDT struct ({type, size, indices, values}) returned from the server into a typed SparseVector. Also extends dataframe.cpp's value printer to handle INT8 and STRUCT Arrow types so vector columns render in show().

Key Implementation Details

• SparseVector / DenseVector (src/ml/linalg/vectors/sparse_vector.h, src/ml/linalg/vectors/dense_vector.h)
• Vector UDT decoding (src/types.h): Added a Row::get<SparseVector>() specialization.
• Arrow rendering (src/dataframe.cpp): Added INT8 and STRUCT cases to arrayValueToString.

API Changes

• New public types: SparseVector, DenseVector.
• New Row::get<SparseVector>(column_name) new template instantiation.

Testing

• New Integration Test: SparkIntegrationTest.SparkVector (tests/spark/ml/linalg/vectors/sparse_vector.cpp) - verifies the decode + norm path with a unit test that constructs a Row mirroring Spark's vector UDT and asserts the norms.

Why is this change necessary?

DataFrames containing ML feature columns (e.g. output of HashingTF transformer) can be consumed. Sparse vectors are the first piece; DenseVector decoding and other linalg ops will follow.

User-Facing Changes

Does this introduce a user-facing change? Yes.

#include "ml/linalg/vectors/sparse_vector.h"

// Given a DataFrame with a Spark ML vector column "features":
auto rows = df.collect();
SparseVector v = rows[0]->get<SparseVector>("features");

std::cout << "nnz="   << v.numNonzeros() << "\n"
          << "L1="    << v.norm(1)       << "\n"
          << "L2="    << v.norm(2)       << "\n"
          << "Linf="  << v.norm(3)       << "\n"
          << "argmax="<< v.argmax()      << "\n";