refactor: move internal headers to numpy/detail/

Create numpy/detail/ subdirectory so the directory layout itself enforces
the public/internal API boundary — no navigating is needed to know what is
off-limits to external callers.

Files moved (git mv):
  numpy/npy_math_float.h  → numpy/detail/npy_math_float.h
  numpy/svml_bridge.h     → numpy/detail/svml_bridge.h
  numpy/blas_bridge.h     → numpy/detail/blas_bridge.h
  numpy/avx512_loops.h    → numpy/detail/avx512_loops.h

Include paths updated:
  core.h: "svml_bridge.h" → "detail/svml_bridge.h"  (and blas, avx512)
  svml_bridge.h: "npy_math_float.h" unchanged (same directory)

Box-comment wrong-path examples updated to show new paths (e.g.
  ✗ #include "numpy/detail/svml_bridge.h" ← compile error)

Installed layout (dpkg -L numpycpp-dev):
  /usr/include/numpycpp/numpy/core.h          ← public
  /usr/include/numpycpp/numpy/linalg.h        ← public
  /usr/include/numpycpp/numpy/einsum.h        ← public
  /usr/include/numpycpp/numpy/detail/*.h      ← all internal, locked

754 tests pass, 0 regressions.

Author: peng.li24

numpy/core.h

@@ -9,10 +9,10 @@
 //
 //  The four internal headers pulled in below are LOCKED behind
 //  NUMPYCPP_INTERNAL_INCLUDE and will cause a #error if included directly:
-//    • svml_bridge.h   — SVML/npy scalar bridge  (x86_64 + Linux)
-//    • blas_bridge.h   — OpenBLAS ILP64 bridge   (x86_64 + Linux)
-//    • npy_math_float.h— float32 poly kernels    (numpy internal constants)
-//    • avx512_loops.h  — AVX-512 specializations (requires AVX-512F CPU)
+//    • detail/svml_bridge.h    — SVML/npy scalar bridge  (x86_64 + Linux)
+//    • detail/blas_bridge.h    — OpenBLAS ILP64 bridge   (x86_64 + Linux)
+//    • detail/npy_math_float.h — float32 poly kernels    (numpy internal constants)
+//    • detail/avx512_loops.h   — AVX-512 specializations (requires AVX-512F CPU)
 //
 //  All functions operate on raw pointers + sizes.
 //  Each function is annotated with its Python numpy equivalent,
@@ -41,9 +41,9 @@
 // The macro below is the compile-time lock; it is #undef-ed at the end of this
 // file so it cannot "leak" into translation units that include core.h.
 #define NUMPYCPP_INTERNAL_INCLUDE
-#include "svml_bridge.h"   // numpy::detail::{exp,log,sin,...}_f32/f64 — SVML/npy
-#include "blas_bridge.h"   // numpy::detail::blas_ops<T> — OpenBLAS ILP64
-// avx512_loops.h included at namespace-close (line ~1004), also guarded.
+#include "detail/svml_bridge.h"   // numpy::detail::{exp,log,sin,...}_f32/f64
+#include "detail/blas_bridge.h"   // numpy::detail::blas_ops<T> — OpenBLAS ILP64
+// detail/avx512_loops.h included at namespace-close below, also guarded.
 
 namespace numpy {
 
@@ -1021,7 +1021,7 @@ inline void norm_axis(const T* src, T* dst, const ptrdiff_t* shape, int ndim, in
 // AVX-512 wide-loop template specializations (0 ULP, ~8-16x faster than scalar)
 // Must be included inside namespace numpy after all primary templates.
 // ============================================================================
-#include "avx512_loops.h"
+#include "detail/avx512_loops.h"
 
 // Release the internal-include lock so it does not pollute the includer's TU.
 #undef NUMPYCPP_INTERNAL_INCLUDE

numpy/avx512_loops.h numpy/detail/avx512_loops.hnumpy/avx512_loops.h renamed to numpy/detail/avx512_loops.h

@@ -5,7 +5,7 @@
 // ║  generic loops in core.h.  It is x86_64 + AVX-512F specific and must   ║
 // ║  be included INSIDE namespace numpy at the end of core.h — nowhere else.║
 // ║                                                                          ║
-// ║  ✗  #include "numpy/avx512_loops.h"     ← compile error                ║
+// ║  ✗  #include "numpy/detail/avx512_loops.h"     ← compile error                ║
 // ║  ✓  #include "numpy/core.h"             ← only correct entry point      ║
 // ╚══════════════════════════════════════════════════════════════════════════╝
 //

numpy/blas_bridge.h numpy/detail/blas_bridge.hnumpy/blas_bridge.h renamed to numpy/detail/blas_bridge.h

@@ -5,7 +5,7 @@
 // ║  All symbols live in numpy::detail — an implementation namespace that   ║
 // ║  external code must never reference.                                     ║
 // ║                                                                          ║
-// ║  ✗  #include "numpy/blas_bridge.h"      ← compile error                ║
+// ║  ✗  #include "numpy/detail/blas_bridge.h"      ← compile error                ║
 // ║  ✗  numpy::detail::blas_sdot(...)       ← undefined behaviour          ║
 // ║  ✓  #include "numpy/core.h"             ← only correct entry point      ║
 // ║  ✓  numpy::dot(a, b, n)                 ← public API                    ║

numpy/npy_math_float.h numpy/detail/npy_math_float.hnumpy/npy_math_float.h renamed to numpy/detail/npy_math_float.h

@@ -5,7 +5,7 @@
 // ║  are tied to numpy's internal SIMD constants.  The API is UNSTABLE and  ║
 // ║  subject to change without notice.                                       ║
 // ║                                                                          ║
-// ║  ✗  #include "numpy/npy_math_float.h"   ← compile error                ║
+// ║  ✗  #include "numpy/detail/npy_math_float.h"   ← compile error                ║
 // ║  ✓  #include "numpy/core.h"             ← only correct entry point      ║
 // ╚══════════════════════════════════════════════════════════════════════════╝
 

numpy/svml_bridge.h numpy/detail/svml_bridge.hnumpy/svml_bridge.h renamed to numpy/detail/svml_bridge.h

@@ -6,7 +6,7 @@
 // ║  All symbols live in numpy::detail — an implementation namespace that   ║
 // ║  external code must never reference.                                     ║
 // ║                                                                          ║
-// ║  ✗  #include "numpy/svml_bridge.h"      ← compile error                ║
+// ║  ✗  #include "numpy/detail/svml_bridge.h"      ← compile error                ║
 // ║  ✗  numpy::detail::exp_svml_f64(x)      ← undefined behaviour          ║
 // ║  ✓  #include "numpy/core.h"             ← only correct entry point      ║
 // ║  ✓  numpy::exp(src, dst, n)             ← public API                    ║