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)	4.48 ✅	5.87 ✅	0.75 🟡	4.42 ✅	3.33 ✅	1.66 ✅	2.80 ✅
1-D strided a[::2]	3.80 ✅	4.13 ✅	1.29 ✅	0.57 🟡	0.57 🟡	2.04 ✅	1.54 ✅
1-D reversed a[::-1]	4.67 ✅	4.38 ✅	0.69 🟡	0.53 🟡	0.48 🟠	1.65 ✅	1.35 ✅
array + scalar	5.52 ✅	4.46 ✅	0.73 🟡	4.60 ✅	3.04 ✅	1.79 ✅	2.77 ✅
scalar + array	5.55 ✅	8.52 ✅	0.74 🟡	4.63 ✅	3.50 ✅	1.69 ✅	3.14 ✅
mixed C + F	2.86 ✅	3.89 ✅	0.77 🟡	0.93 🟡	0.74 🟡	1.77 ✅	1.48 ✅
mixed C + T	3.18 ✅	0.89 🟡	0.77 🟡	0.91 🟡	0.75 🟡	2.40 ✅	1.23 ✅
binary broadcast +row(1,C)	5.75 ✅	3.01 ✅	0.74 🟡	4.97 ✅	3.84 ✅	1.71 ✅	2.73 ✅
binary broadcast +col(R,1)	5.55 ✅	3.06 ✅	0.69 🟡	5.42 ✅	6.77 ✅	2.60 ✅	3.23 ✅
col-broadcast unary (inner stride-0)	5.12 ✅	1.98 ✅	0.95 🟡	2.27 ✅	3.42 ✅	5.45 ✅	2.72 ✅

Worst 12 cells

key	NumSharp ms	NumPy ms	ratio
1d_rev	i64	3.6240	1.7425	0.48 🟠
1d_rev	i32	1.8063	0.9642	0.53 🟡
1d_strided	i64	1.8131	1.0274	0.57 🟡
1d_strided	i32	0.9047	0.5148	0.57 🟡
1d_rev	f16	9.7203	6.6619	0.69 🟡
bcast_col	f16	9.5953	6.6540	0.69 🟡
scalar_rhs	f16	9.0203	6.5493	0.73 🟡
bcast_row	f16	9.1483	6.7400	0.74 🟡
scalar_lhs	f16	8.9600	6.6276	0.74 🟡
mix_C_F	i64	3.9903	2.9637	0.74 🟡
1d_C	f16	8.9710	6.6916	0.75 🟡
mix_C_T	i64	3.7170	2.7903	0.75 🟡

60 comparable cells.