address PR #562 review: include hygiene, K_ re-init, gemm_batched congruence checks

- einsum/tiledarray.h, tile_op/batched_contract_reduce.h, pmap/slabbed_pmap.h:
  include what is used (<cstdlib>, <string_view>, util/vector.h, <memory>,
  <utility>) instead of relying on transitive includes
- cont_engine.h: re-initialize K_ in init_distribution_general() (defensive;
  engines are single-use, but mirrors the n_slabs_ reset)
- sparse_shape.h: gemm_batched() now TA_ASSERTs that the argument ranks match
  the folded gemm ranks plus the fused modes and that the fused and contracted
  mode extents of the two shapes are congruent (the batched analogue of the
  checks GemmHelper::compute_matrix_sizes performs for plain gemm)

Author: evaleev

src/TiledArray/einsum/tiledarray.h

@@ -14,7 +14,9 @@
 
 #include <madness/world/thread.h>
 
+#include <cstdlib>
 #include <optional>
+#include <string_view>
 
 namespace TiledArray {
 enum struct DeNest { True, False };

src/TiledArray/expressions/cont_engine.h

@@ -793,6 +793,7 @@ class ContEngine : public BinaryEngine<Derived> {
     // Compute the slab count and the fused sizes of the per-slab contraction
     size_type M = 1ul, m = 1ul, N = 1ul, n = 1ul;
     n_slabs_ = 1ul;
+    K_ = 1ul;
     for (unsigned int i = 0u; i < nh; ++i) n_slabs_ *= left_tiles_size[i];
     for (unsigned int i = nh; i < nh + neA; ++i) {
       M *= left_tiles_size[i];

src/TiledArray/pmap/slabbed_pmap.h

@@ -25,6 +25,9 @@
 
 #include <TiledArray/pmap/pmap.h>
 
+#include <memory>
+#include <utility>
+
 namespace TiledArray {
 namespace detail {
 

src/TiledArray/sparse_shape.h

@@ -1719,6 +1719,20 @@ class SparseShape {
     using integer = TiledArray::math::blas::integer;
     const auto* left_extent = tile_norms_.range().extent_data();
     const auto* right_extent = other.tile_norms_.range().extent_data();
+
+    // check that the ranks match the folded gemm ranks plus the fused modes,
+    // and that the fused and contracted mode extents of the two shapes are
+    // congruent
+    TA_ASSERT(tile_norms_.range().rank() == nfused + gemm_helper.left_rank());
+    TA_ASSERT(other.tile_norms_.range().rank() ==
+              nfused + gemm_helper.right_rank());
+    for (unsigned int d = 0u; d < nfused; ++d)
+      TA_ASSERT(left_extent[d] == right_extent[d]);
+    for (unsigned int i = gemm_helper.left_inner_begin(),
+                      j = gemm_helper.right_inner_begin();
+         i < gemm_helper.left_inner_end(); ++i, ++j)
+      TA_ASSERT(left_extent[nfused + i] == right_extent[nfused + j]);
+
     integer H = 1, M = 1, N = 1, K = 1;
     for (unsigned int d = 0u; d < nfused; ++d) H *= left_extent[d];
     for (unsigned int i = gemm_helper.left_outer_begin();

src/TiledArray/tile_op/batched_contract_reduce.h

@@ -24,6 +24,7 @@
 #define TILEDARRAY_TILE_OP_BATCHED_CONTRACT_REDUCE_H__INCLUDED
 
 #include <TiledArray/tile_op/contract_reduce.h>
+#include <TiledArray/util/vector.h>
 
 namespace TiledArray {
 namespace detail {