[llvm] [SLP] Account for GEP pointer-chain cost when root scalars feed load/store indices (PR #192726)

[Ryan Buchner via llvm-commits]

bababuck wrote:

On line 4749, there's a typo in the comment which could get wrapped in with this MR: (`reaplced` -> `replaced`)
```
  /// A list of GEPs which can be reaplced by scalar GEPs instead of
  /// extractelement instructions.
  SmallPtrSet<Value *, 4> ExternalUsesAsOriginalScalar;
```

https://github.com/llvm/llvm-project/pull/192726