Add zero-dependency native C extension for LSI acceleration

[cardmagic]

Summary

Replace the rb-gsl dependency with a self-contained C extension that implements Vector, Matrix, and Jacobi SVD operations. This eliminates the need for users to install external libraries (like libgsl) while providing significant performance improvements.

• Add native C extension (~850 lines) with Vector, Matrix, and SVD implementations
• Port existing Ruby Jacobi SVD algorithm to C for consistent results
• Auto-detect backend: native extension → pure Ruby fallback
• Remove all GSL-related code and dependencies
• Update benchmarks and documentation

Performance

Documents	build_index	Overall Speedup
5	7x	2.6x
10	25x	4.6x
15	112x	14.5x
20	385x	48.7x

Detailed benchmark (20 documents)

Operation            Pure Ruby     Native C      Speedup
----------------------------------------------------------
build_index            0.5540       0.0014       384.5x
classify               0.0190       0.0060         3.2x
search                 0.0145       0.0037         3.9x
find_related           0.0098       0.0011         8.6x
----------------------------------------------------------
TOTAL                  0.5973       0.0123        48.7x

Test Plan

• All 113 tests pass with native C extension
• All 91 tests pass with pure Ruby fallback (NATIVE_VECTOR=true)
• Benchmark comparison shows expected speedups
• No compiler warnings in C code
• No Ruby warnings during test runs

Files Changed

New:

• ext/classifier/ - C extension source files
• test/linalg/native_ext_test.rb - Unit tests for native extension

Modified:

• classifier.gemspec - Add extension configuration
• lib/classifier/lsi.rb - Backend detection logic
• README.md / CLAUDE.md - Documentation updates

Removed:

• lib/classifier/extensions/vector_serialize.rb - GSL serialization (no longer needed)

Closes #87