Variable not in the first block in the function

In MLIR or LLVMIR, allocas are not required to be in the beginning of a function. While in SPIR-V, according to https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html#_logical_layout_of_a_module, "All OpVariable instructions in a function must be in the first block in the function."
We expects the translator to move all allocas to the first block of a function.
We noticed incorrect result when there exist alloca not in the first block in the function.

Reproducer: [reproducer.zip](https://github.com/KhronosGroup/SPIRV-LLVM-Translator/files/10970669/reproducer.zip)

In the reproducer, there are 2 LLVM IR files (`good.ll` and `bad.ll`), the only difference is the location of one alloca instruction.
We can see from `bad.spt` (created from `bad.ll`) at line 1279 that `4 Variable 249 250 7` is placed not in the first block in the function.

We can see from below that using `good.ll` in the `script`, it generates the expected result. While using `bad.ll` in the `script`, it hits an assertion.
```
bash-4.4$ ./script good
Test passed.
bash-4.4$ ./script bad
linear-sub_group.cpp.tmp.out: linear-sub_group.cpp:51: int main(int, char **): Assertion `output[idx] == idx' failed.
./script: line 42: 3238678 Aborted (core dumped) "env" "ONEAPI_DEVICE_SELECTOR=opencl:cpu" "./linear-sub_group.cpp.tmp.out"
```

Start by examining the supplied good.ll and bad.ll files and running the script with each reproducer. Compare the generated bad.spt around line 1279 and the assertion output; done means the alloca-derived OpVariable is placed in the function's first block and the bad case no longer produces the incorrect result.

Written by the indexing model from the issue text.