Fix ARM64 SME direct-kernel build and dispatch

[moluopro]

• Add the missing CMake compile definitions for ARM64 SME direct-kernel variants to match the existing kernel/Makefile.L3 rules.
• Restrict the STRMM direct fast path in interface/trsm.c to TRMM builds so STRSM cannot dispatch to STRMM direct kernels.
• Treat real-valued CblasConjNoTrans and CblasConjTrans as eligible for the corresponding STRMM direct paths.
• Add an alpha == 0 fast zeroing path in strmm_direct_arm64_sme1.c.
• Avoid signed overflow in the M * N * K size heuristic in sgemm_direct_performant.c, and skip the direct path when N <= 0 or K <= 0.
• Clean up the Windows/Clang condition for ARMV9SME and fix the surrounding indentation in cmake/cc.cmake.
• Normalize the relevant SME source files under kernel/arm64 from CRLF to LF line endings.

[moluopro]

This follows up on #5889.

While working on the ARM64 SME TRMM direct path, I found a few related build and dispatch issues and fixed them here as well.

[moluopro]

The jenkins/pr-merge pipeline seems to have failed on the last few PRs as well. Looks like Jenkins needs a good kick or two.

[martin-frbg]

The jenkins/pr-merge pipeline seems to have failed on the last few PRs as well. Looks like Jenkins needs a good kick or two.

Seems the Z arch build cannot find its base Ubuntu docker file recently. I will try to reach out to OSUOSL if the problem persists, but they're a small volunteer group too.