Add a WASM SIMD128 4x4 DTRMM microkernel

[jjerphan]

Summary

Follow-up to #5983 / #5680 / #4023: replace the generic scalar 4×4 DTRMM kernel used by WASM128_GENERIC with a WASM SIMD128 microkernel. Double only.

• Generic DTRMM was the remaining Level-3 gap versus DGEMM after Add a 4x4 WASM SIMD128 GEMM microkernel #5983. DTRMMKERNEL now uses kernel/wasm/trmmkernel_4x4_wasm128.c; packing stays gemm_{n,t}copy_4.
• STRMM stays ../generic/trmmkernel_4x4.c. Single-precision TRMM already auto-vectorizes, and a shared S+D SIMD kernel previously slowed SGEMM.
• Relaxed madd is used in the TRMM inner k-loop only (same as GEMM in Add a 4x4 WASM SIMD128 GEMM microkernel #5983). ifndef DTRMMKERNEL in kernel/wasm/KERNEL already keeps the target file from being overwritten.

Benchmarks

Node / Emscripten, TARGET=WASM128_GENERIC, USE_THREAD=0, COMMON_OPT=-O2. Same machine, 5 warmup + 10 timed samples (median). Speedup > 1 means this branch is faster than develop.

Geomean vs develop: DTRMM 2.42x. STRMM 1.02x, SGEMM 1.03x, DGEMM 1.03x (unchanged within noise).

op	n	develop MFLOPS	SIMD MFLOPS	speedup
dtrmm	32	5891.7	17215.2	2.922
dtrmm	64	6262.5	19206.8	3.067
dtrmm	128	7241.3	20610.8	2.846
dtrmm	256	11069.7	21880.9	1.977
dtrmm	512	14937.4	24349.6	1.630

Test plan

• Node CBLAS odd-size GEMM/TRSM/TRMM check
• utest under node (106/106)
• CBLAS ctest L1/L2/L3 (x{s,d,c,z}cblat{1,2,3}) under node, including STRMM/DTRMM