[llvm] [SPIR-V] Add implementation of G_SPLAT_VECTOR opcode and fix invalid types processing (PR #84766)

[Thorsten Schütt via llvm-commits]

tschuett wrote:

The tests in the PR show:
```
(<vscale x 2 x s1>) = G_SPLAT_VECTOR [[EVEC]](s1)
```
There is now a `vscale` for G_SPLAT_VECTOR.

Great. Sorry for the confusion.

https://github.com/llvm/llvm-project/pull/84766