[llvm] [AArch64][SLP] Add NFC test cases for floating point reductions (PR #106507)

[Sushant Gokhale via llvm-commits]

sushgokh wrote:

> Can you explain what it is you plan to change? All the tests are `fast` and it might be something that needs to dependent on that.

The reduction cost for @llvm.vector.reduce.fadd with 'fast'

https://github.com/llvm/llvm-project/pull/106507