S1-30: SSM_CONV_BACK CPU kernel

[dillon-blake]

Stacked on #23 (S1-29b).

d_sx[i2 + i0, i1, i3] += dy[i1, i2, i3] * c[i0, i1]

The forward is a depthwise causal convolution over a sliding window, so its gradient scatters each output's gradient back across the d_conv inputs that produced it. Written as a scatter rather than the equivalent gather (d_sx[j] = sum_t dy[t]*c[j-t], with its two-sided bounds on t) because the scatter needs no boundary arithmetic at all — every (i2, i0) pair lands in range by construction.

The leading d_conv - 1 columns of sx are the carried convolution state and receive a gradient like any other input. Dropping them would silently truncate the gradient at every sequence boundary.

Threaded by (row, sequence): each (i1, i3) owns its own d_sx column — no atomics, no barrier, fixed accumulation order.

Under sum(out) this test was vacuous, and I only found out by mutating the kernel

sum(out) makes the incoming gradient all-ones, and the scatter is dy * c — so with dy == 1, a kernel that ignores dy entirely and scatters c alone produces exactly the same answer.

mutation + objective	result
ignore the incoming gradient, default sum(out)	not caught at all
ignore the incoming gradient, weighted grad_loss	MAA 0.84

Third instance of the same structural trap (SOFT_MAX in S1-34, MUL_MAT_ID in S1-27). The weighted grad_loss hook is now the default assumption for any new VJP, not an afterthought.

Verification

max_maa_err = 1e-2, measured with numbers on both sides:

	MAA
worst FD noise, 25 runs	2.2e-3 (on the widest shapes)
ignore the incoming gradient	0.84
shift the scatter window by 1	0.52
drop the last conv tap	0.47

Three mutations injected, three caught.

Also checked exactly — against a naive double-precision reference and with a transposed grad (nb[0] != 4), the stride trap that bit both MoE kernels and GLU_BACK. Exact to ~1e-7 in both layouts; this kernel reads dy through src0->nb[0] from the start.

54 grad cases now execute. They previously reported OK while requesting no gradients at all.