Add a 4x4 WASM SIMD128 GEMM microkernel

[jjerphan]

Summary

Follow-up to #5680 / #4023: replace the 2x2 shuffle GEMM for WASM128_GENERIC with a 4x4 outer-product kernel so f32x4 / f64x2 loads match packing width.

• SGEMM/DGEMM use gemmkernel_4x4_wasm128.c with matching gemm_{n,t}copy_4 and generic TRMM 4x4.
• KERNEL TRMM defaults are wrapped in ifndef so KERNEL.WASM128_GENERIC packing width is not overwritten.
• -mrelaxed-simd plus relaxed_madd in the GEMM inner loop only; DOT/AXPY stay IEEE mul+add in intrin_wasm.h.
• Separate commit gates leftover WASM TRSM call counters behind OPENBLAS_WASM_TRSM_PROFILE.

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: SGEMM 1.44x, DGEMM 1.82x, SSYRK 1.46x, DSYRK 1.73x, STRSM 1.33x, DTRSM 1.39x. Level-1/2 kernels are unchanged within noise.

op	n	develop MFLOPS	4x4 MFLOPS	speedup
sgemm	32	14141.8	27612.0	1.953
sgemm	64	18077.5	26887.8	1.487
sgemm	128	20662.3	25570.7	1.238
sgemm	256	20922.9	28427.7	1.359
sgemm	512	22546.2	28625.6	1.270
dgemm	32	9728.9	17933.1	1.843
dgemm	64	10937.9	19025.1	1.739
dgemm	128	11355.1	20954.0	1.845
dgemm	256	11948.5	21770.6	1.822
dgemm	512	11893.9	22016.5	1.851
ssyrk	32	9938.9	17646.4	1.775
ssyrk	64	15650.7	23219.1	1.484
ssyrk	128	19700.5	26626.8	1.352
ssyrk	256	21511.9	27573.4	1.282
dsyrk	32	7782.4	12758.0	1.639
dsyrk	64	9920.3	17041.1	1.718
dsyrk	128	10926.3	19363.8	1.772
dsyrk	256	11743.9	21167.8	1.802
strsm	32	6908.7	10204.4	1.477
strsm	64	10845.4	15067.1	1.389
strsm	128	15141.3	19255.0	1.272
strsm	256	19006.7	22809.6	1.200
dtrsm	32	5288.2	6726.3	1.272
dtrsm	64	7730.9	9780.4	1.265
dtrsm	128	9644.2	13868.7	1.438
dtrsm	256	10760.1	17370.6	1.614

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 SGEMM/DGEMM, STRMM/DTRMM, STRSM/DTRSM, SSYRK/DSYRK