CK: Remove duplicate bpreshuffle MX pipeline header (~1,148 lines)

[AviralGoelAMD]

Summary

Remove blockwise_gemm_pipeline_xdlops_v3_mx_bpreshuffle.hpp, a copy-paste duplicate of blockwise_gemm_pipeline_xdlops_b_preshuffle_mx_moe_v3.hpp. The files are code-identical (1,148 lines each, only the class name differs). This PR keeps the canonical b_preshuffle_mx_moe_v3 version and updates the mx_bpreshuffle_selector to use it.

Changes

File	Action	Detail
blockwise_gemm_pipeline_xdlops_v3_mx_bpreshuffle.hpp	Deleted	1,148-line duplicate
blockwise_gemm_pipeline_xdlops_mx_bpreshuffle_selector.hpp	Modified	Updated include + class name reference

Verification

• Grep confirms zero remaining references to the deleted file or its class name (BlockwiseGemmXdlops_pipeline_v3_mx_bprehuffle)
• The two BlockGemmMXBPreshufflePipeline_Selector overloads (different template signatures) are included in separate TUs — no ODR risk
• Also fixes a typo in the old class name (bprehuffle -> bpreshuffle)

Cumulative Cleanup Stats

PR	Category	Net Lines
#6300	Dead #if 0 blocks	-2,648
#6302	Commented-out code	-2,861
#6303	Orphaned files	-3,886
#6323	Gemm quant test dedup	-12,106
#6324	Contraction example dedup	-8,220
#6325	Contraction instance dedup	-6,511
This PR	Duplicate pipeline header	-1,148
Total		-37,380

[AviralGoelAMD]

Closing: deeper analysis revealed the two files have diverged beyond class names. PRs #3098 and #2455 modified the kept file but not the deleted file, introducing real code differences (static_for vs static_ford loop structure, scale offset calculation patterns). Not safe to consolidate without verifying which variant is correct for each kernel path.