[llvm] [VPlan][NFC] Simplify reverse access pattern detection in EVL vectorization (PR #199510)

[Mel Chen via llvm-commits]

Mel-Chen wrote:

> That's interesting! I'll check the test suite results for the new implementation tomorrow. Thanks for your report.

I tested llvm-test-suite using the following config:
```
-O3 -mcpu=sifive-x280 -save-temps=obj
```
For the new implementation, the differences only come from commit hash. 
``` diff
diff --git a/MultiSource/Benchmarks/7zip/CMakeFiles/7zip-benchmark.dir/CPP/Common/UTFConvert.s b/MultiSource/Benchmarks/7zip/CMakeFiles/7zip-benchmark.dir/CPP/Common/UTFConvert.s
index f60bf0d..5bc7a15 100644
--- a/MultiSource/Benchmarks/7zip/CMakeFiles/7zip-benchmark.dir/CPP/Common/UTFConvert.s
+++ b/MultiSource/Benchmarks/7zip/CMakeFiles/7zip-benchmark.dir/CPP/Common/UTFConvert.s
@@ -843,6 +843,6 @@ _ZL11kUtf8Limits:
        .ascii  "\300\340\360\370\374"
        .size   _ZL11kUtf8Limits, 5

-       .ident  "clang version 23.0.0git (git at github.com:Mel-Chen/llvm-project.git 5eae7dc2ac7d8dde06a943771b792aeb91c31f57)"
+       .ident  "clang version 23.0.0git (git at github.com:Mel-Chen/llvm-project.git 1f7de98c96f7bf485a41d01a0e8aaee7a1d6dfc8)"
        .section        ".note.GNU-stack","", at progbits
        .addrsig
```
This patch is currently confirmed to be NFC for the llvm-test-suite. The previous version probably introduced unexpected transformations via simplifyRecipes, I think.

https://github.com/llvm/llvm-project/pull/199510