Operand & broadcast layouts — 1-D / scalar / mixed-operand / broadcast (NumSharp vs NumPy 2.4.2)

The layout classes the per-operand layout grid (benchmark/layout) can't express. ratio = NumPy_ms / NumSharp_ms — >1.0 = NumSharp faster. ✅≥1.0 🟡≥0.5 🟠≥0.2 🔴<0.2. 1M elements, best-of-3.

case	f64	f32	f16	i32	i64	c128	geomean
1-D contiguous (a+a)	2.41 ✅	2.16 ✅	0.63 🟡	2.13 ✅	2.45 ✅	2.35 ✅	1.85 ✅
1-D strided a[::2]	1.84 ✅	1.46 ✅	0.53 🟡	1.51 ✅	1.81 ✅	1.93 ✅	1.40 ✅
1-D reversed a[::-1]	2.49 ✅	1.95 ✅	0.56 🟡	2.14 ✅	2.29 ✅	2.22 ✅	1.76 ✅
array + scalar	2.72 ✅	1.93 ✅	0.63 🟡	1.91 ✅	2.26 ✅	2.64 ✅	1.83 ✅
scalar + array	2.40 ✅	1.98 ✅	0.64 🟡	2.01 ✅	2.19 ✅	2.69 ✅	1.82 ✅
mixed C + F	2.29 ✅	2.02 ✅	0.62 🟡	2.02 ✅	2.09 ✅	1.87 ✅	1.68 ✅
mixed C + T	2.53 ✅	2.04 ✅	0.62 🟡	2.01 ✅	2.33 ✅	2.30 ✅	1.80 ✅
binary broadcast +row(1,C)	2.71 ✅	2.09 ✅	0.63 🟡	1.95 ✅	2.52 ✅	2.58 ✅	1.89 ✅
binary broadcast +col(R,1)	2.61 ✅	2.02 ✅	0.56 🟡	2.07 ✅	2.54 ✅	2.96 ✅	1.89 ✅
col-broadcast unary (inner stride-0)	2.54 ✅	1.56 ✅	0.88 🟡	1.65 ✅	2.64 ✅	6.22 ✅	2.13 ✅

Worst 12 cells

key	NumSharp ms	NumPy ms	ratio
1d_strided	f16	2.6261	1.3926	0.53 🟡
1d_rev	f16	5.2838	2.9653	0.56 🟡
bcast_col	f16	5.2887	2.9837	0.56 🟡
mix_C_T	f16	5.3031	3.2873	0.62 🟡
mix_C_F	f16	5.3153	3.3015	0.62 🟡
1d_C	f16	4.7456	2.9682	0.63 🟡
bcast_row	f16	4.7695	2.9999	0.63 🟡
scalar_rhs	f16	4.7058	2.9602	0.63 🟡
scalar_lhs	f16	4.6383	2.9742	0.64 🟡
colbcast_unary	f16	0.4753	0.4173	0.88 🟡
1d_strided	f32	0.2366	0.3444	1.46 ✅
1d_strided	i32	0.2433	0.3676	1.51 ✅

60 comparable cells.